The Autonomous Robotic Collaboration Lab

Worcester Polytechnic Institute

People

Who's working in the ARC lab

Welcome

News

  • We've released a toolkit for ROS communication over low-bandwidth and high-latency connections. (wiki | code)
  • The ARC lab is recruiting both graduate and undergraduate students. Contact dberenson [at] cs.wpi.edu if you're interested!
  • Join the ARC mailing list to hear about the latest news. Send an email to dberenson [at] cs.wpi.edu to subscribe.

Research Interests

Our research focuses on creating algorithms that allow robots to interact with the world and collaborate efficiently with people. These general-purpose motion planning and manipulation algorithms can be applied to robots that work in homes, factories, and operating rooms. We are interested in all aspects of algorithm development; including creating efficient algorithms, proving their theoretical properties, validating them on real-world robots and problems, integrating them with sensing and higher-level reasoning, and distributing them to open-source communities. Our lab draws on ideas in search, optimization, control theory, and topology to develop these planning algorithms and to prove their properties. We also seek to develop algorithms which can generalize to many types of practical tasks and application areas.


Press

  • Research in the ARC lab featured in an article about the future of service robots in Robotics Industry Association Industry Insights(link)

  • WPI robotics program and our work on human-robot collaboration featured in Robotics Business Review (link)

  • New York Times article featuring the two DARPA Robotics Challenge Teams at WPI (link)

  • The ARC Lab leads WPI's contribution to a multi-university Track A DARPA Robotics Challenge Team (link)

  • The Worcester Telegram and Gazette covers our work on the DARPA Robotics Challenge (link)

  • Worcester News Tonight features Archie (link)





For technical information relevant to working in the lab, please see arc.wpi.edu/wiki.

Human-Robot Collaboration for Manipulation
Recent hardware developments in robotics have made human-robot collaboration physically possible, but robots still require new algorithms to ensure safety, efficiency, and fluency when working with people.
Expand

This project addresses a large space of tasks that cannot be fully automated because of either the limitations of current algorithms or prohibitive cost and set-up time. Such tasks generally require humans to collaborate in close proximity and adapt to each other’s decisions and motions. This project explores accomplishing these tasks through human-robot collaboration.

Recent hardware developments in robotics have made human-robot collaboration physically possible, but robots still require new algorithms to ensure safety, efficiency, and fluency when working with people. Creating such algorithms is difficult because there can be high uncertainty in what a person is going to do and how they are going to do it. This project explores the integration of reasoning about how a person moves and how he or she makes decisions into a robot motion planning and decision-making framework. The research centers on the development of new algorithmic frameworks for modeling, simulating, and planning for human-robot collaboration, which requires advances in robot training, task modeling, human motion understanding, high-dimensional motion planning with uncertainty, and metrics to assess human-robot joint action.





  • Human-Robot Collaborative Manipulation Planning Using Early Prediction of Human Motion
    Jim Mainprice and Dmitry Berenson
    IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), November, 2013
    PDF | Video

  • User-Guided Manipulation for the DARPA Robotics Challenge
    We seek to create a user-guided manipulation framework for High Degree-of-Freedom robots operating in environments with limited communication to apply to the DARPA Robotics Challenge.
    Expand

    We seek to create a user-guided manipulation framework for High Degree-of-Freedom robots operating in environments with limited communication to apply to the DARPA Robotics Challenge. Our approach consists of three elements: (1) a user-guided perception interface which assists the user to provide task level commands to the robot, (2) planning algorithms that autonomously generate robot motion while obeying relevant constraints, and (3) a trajectory execution and monitoring system. We are prototyping our work on the PR2 but are applying it for the DRC on the Hubo humanoid robot.

    We are part of a multi-university Track A DARPA Robotics Challenge Team led by Drexel University.

    See a description of the challenge here.

    This project is joint work with the Interaction Lab (Prof. Sonia Chernova) and the HIVE lab (Prof. Rob Lindeman) at WPI.



  • Toward A User-Guided Manipulation Framework for High-DOF Robots with Limited Communication
    Nicholas Alunni, Calder Phillips-Grafflin, Halit Bener Suay, Daniel Lofaro, Dmitry Berenson, Sonia Chernova, Robert W. Lindeman, Paul Oh
    IEEE International Conference on Technologies for Practical Robot Applications (TePRA), April, 2013
    PDF | Video

  • From Autonomy to Cooperative Traded Control of Humanoid Manipulation Tasks with Unreliable Communication: System Design and Lessons Learned
    Jim Mainprice, Calder Phillips-Grafflin, Halit Bener Suay, Nicholas Alunni, Daniel Lofaro, Dmitry Berenson, Sonia Chernova, Robert W. Lindeman, and Paul Oh
    IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), November, 2014
    PDF | Video

  • Toward a user-guided manipulation framework for high-DOF robots with limited communication
    Calder Phillips-Grafflin, Nicholas Alunni, Halit Bener Suay, Jim Mainprice, Daniel Lofaro, Dmitry Berenson, Sonia Chernova, Robert W. Lindeman, and Paul Oh
    Journal of Intelligent Service Robotics, Vol. 7, No. 3, pp. 121–131, July 2014
    PDF

  • Combining Learning and Manipulation Planning
    Learning algorithms are good at quickly producing solutions to problems that have been encountered before, but they have difficulty generalizing to new situations. Planning algorithms are good at generalizing to new situations but at significant (and sometimes prohibitive) computational cost. This project seeks to combine the best of both approaches.
    Expand

    The goal of combining learning and planning is to create algorithms whose performance improves with experience while maintaining generality. Learning algorithms are good at quickly producing solutions to problems that have been encountered before, but they have difficulty generalizing to new situations. Planning algorithms are good at generalizing to new situations but at significant (and sometimes prohibitive) computational cost. Our work seeks to combine the best of both approaches to create a method whose performance improves as it gains more experience while retaining the ability to generalize to new situations. As a first step in this direction, we recently developed a path planning framework that is able to use previous experience by storing and quickly retrieving previously-computed paths and adapting them to the task at hand while also running a planning algorithm in parallel. This framework allows a robot to train itself as it performs tasks without sacrificing generality. This research opens up new directions for autonomous manipulation in many different application areas, including mobile manipulation in the home and surgical robotics, where similar tasks such as opening a refrigerator or tying a suture are performed repeatedly. Many important questions still remain to be answered; including how to divide complex tasks into sub-tasks, how to decide which sub-tasks should be learned and/or planned, and how to use learning to focus planning on the most relevant parts of a high-dimensional search space.



  • A Robot Path Planning Framework that Learns from Experience
    Dmitry Berenson, Pieter Abbeel, and Ken Goldberg
    IEEE International Conference on Robotics and Automation (ICRA), May, 2012.
    Details | PDF | Video

  • Planning for Deformable Objects
    A major frontier for autonomous manipulation lies in interacting with deformable objects, such as cloth and cooking ingredients (in the context of a domestic robot) and thread and tissue (in the context of surgery). This is an extremely under-explored area in autonomous manipulation, mainly because deformable objects are difficult to model and simulate.
    Expand

    We are currently exploring methods that represent elastic deformable objects as voxel grids. This representation is used inside a motion planner to plan low-deformation paths for the a soft/rigid object in a soft/rigid environment. The voxel-based representation requires no simulation to assess the cost of deformation and can be calibrated from real-world experiments.

    We are also exploring methods related to visual-servoing that can guide a deformable object to a nearby target geometry while avoiding excessive stretching and collision of the grippers. This method can also be used to enable collaborative tasks where a human and robot manipulate a deformable object together. The key to our method is a computation of the deformation Jacobian that does not require simulation.



  • Manipulation of Deformable Objects Without Modeling and Simulating Deformation
    Dmitry Berenson
    IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), November, 2013
    PDF | Video

  • A Representation Of Deformable Objects For Motion Planning With No Physical Simulation
    Calder Phillips-Grafflin and Dmitry Berenson
    IEEE International Conference on Robotics and Automation (ICRA), May, 2014.
    PDF | Video

  • Soft Pneumatic Hand
    A soft penumatic hand for the Baxter robot that will enable it to grasp objects with pose, shape, and actuation uncertainty.
    Expand

    Most robotic systems today are rigid, or composed of parts that are hard and unable to flex, morph, or change other physical characteristics. This can be an issue when a rigid robotic hand is trying to grasp something irregularly shaped. These rigid designs need precise control over position, force, and other senses to successfully pick up objects.  Our soft hand design builds on the work of Raphael Deimel and Oliver Brock at TU Berlin in soft pneumatic robot fingers.

     The idea behind a soft robotic hand is to use soft and flexible fingers that can adapt to the objects being picked up.  A soft hand is far more robust in regards to the uncertainty of the objects’ placement, size, and shape.  Although some certainty is still required, less attention needs to be given to the placement of the object in 3D space.  A soft robotic hand can simply morph to the object's shape, and then get a firm grip on it.

     The purpose of this project is to research the development, effectiveness, and practicality of a soft robotic hand that can be used on robots that manipulate complex objects.  The initial goal established was to construct a hand prototype that could be pneumatically controlled to pick up objects of different sizes, shapes, and weights.  This is an ongoing project, however, as there are many areas for expansion.  These include sensor integration, stronger fingers, and a more sophisticated control system.  Now that a proof of concept has been completed, the doors have been opened for a vast range of ways to improve the quality of the hand.  

    This project's hardware and software are completely open-source.

    Source code available on GitHub.

    CAD for the molds and hand.

    Documentation for the hand.







    Path Planning in High-Dimensional Spaces with Cost
    Finding paths in high-dimensional spaces becomes difficult when we wish to optimize the cost of a path in addition to obeying feasibility constraints. While it is possible to find low-cost configurations using sampling, approaches that rely solely on sampling have difficulty navigating cost-space chasms--narrow low-cost regions surrounded by increasing cost.
    Expand

    Finding paths in high-dimensional spaces becomes difficult when we wish to optimize the cost of a path in addition to obeying feasibility constraints. While it is possible to find low-cost configurations using sampling, approaches that rely solely on sampling have difficulty navigating cost-space chasms--narrow low-cost regions surrounded by increasing cost. Such chasms are particularly common in planning for manipulators because many useful cost functions induce narrow or lower-dimensional low-cost areas. We developed the GradienT-RRT algorithm, which combines a sampling-based planner with a local gradient method to bias the search toward lower-cost regions. We found that GradienT-RRT is able to plan paths with lower cost than a previously-presented approach while maintaining better or comparable computation time. This algorithm can be applied to cost functions defined in workspace, task space, and C-space. A primary application is planning with uncertainty, where the cost of a configuration encodes that configuration's probabity of collision. Another application is path imitation, where the cost function can be used to bias the planner toward a demonstrated path while still allowing it the freedom to explore.






  • Addressing Cost-Space Chasms for Manipulation Planning
    Dmitry Berenson, Thierry Simeon, Siddhartha Srinivasa
    IEEE International Conference on Robotics and Automation (ICRA), May, 2011.
    Details | PDF | Video

  • Constrained Manipulation Planning
    Our everyday lives are full of tasks that constrain our movement. Carrying a coffee mug, lifting a heavy object, or sliding a milk jug out of a refrigerator are examples of tasks that involve constraints imposed on our bodies as well as on the manipulated objects. Creating algorithms for general-purpose robots to perform these kinds of tasks also involves computing motions that are subject to multiple simultaneous task constraints.
    Expand
    Overview

    Our everyday lives are full of tasks that constrain our movement. Carrying a coffee mug, lifting a heavy object, or sliding a milk jug out of a refrigerator are examples of tasks that involve constraints imposed on our bodies as well as on the manipulated objects. Creating algorithms for general-purpose robots to perform these kinds of tasks also involves computing motions that are subject to multiple simultaneous task constraints.

    In general, a robot cannot assume arbitrary joint configurations when performing constrained motion. Instead, the robot must move within a manifold embedded in its high-dimensional configuration space that satisfies both the constraints of the task and the limits of the mechanism. Planning such motion is challenging for two reasons: 1) The space of motions is continuous and high-dimensional. 2) Constraints on the robot's motion (imposed by the task and the environment) often make feasible states difficult to find. For instance, constraints on the pose of the robot's hand can restrict the allowable states to an infinitely thin manifold in the configuration space. Such manifolds cannot be searched using standard sampling-based planning techniques.

    My thesis work focused on developing a constrained manipulation planning framework for practical manipulation tasks. The framework has three main components: constraint representation, constraint-satisfaction strategies, and a sampling-based planning strategy. These three components come together to create an efficient and probabilistically-complete manipulation planning algorithm called the Constrained BiDirectional RRT (CBiRRT).


    An implementation of the above framework as the open-source Constrained Manipulation Planning Suite.


    Task Space Regions

    Some of the most common constraints in manipulation planning are constraints on the pose of a robot's end-effector.  They arise in tasks such as reaching to grasp an object, carrying a cup of coffee, or opening a door. These constraints can act on the entire path, such as not spilling a cup of coffee, and/or on the goal of a path, such as reaching to grasp an object. In general a set of constraints for a manipulator's end-effector can consist of an arbitrary number of poses spread in an arbitrary way throughout the task space. However, such a broad representation lacks three fundamental properties that are necessary for sampling-based planning:

    • The set of poses must be easy to specify.
    • Sampling from the set of poses must be efficient.
    • The distance to the set must be fast to compute.

    TSRs describe end-effector pose constraints as volumes in SE(3), the space of rigid spatial transformations. These volumes are particularly useful for manipulation tasks such as reaching to grasp an object, manipulating an object with pose constraints, such as a glass of water, or placing an object onto a surface, such as table. TSRs are intuitive to specify, they can be efficiently sampled, and the distance to a TSR can be evaluated very quickly, making them ideal for sampling-based planners such as CBiRRT.



    Task Space Region Chains for Whole-Body Manipulation

    While TSRs can represent many useful constraints, a single TSR, or even a finite set of TSRs, is sometimes insufficient to capture the pose constraints of a given task.  To describe more complex constraints such as closed chain kinematics and manipulating articulated objects, we use TSR Chains, which are defined by linking a series of TSRs. A TSR Chain can be thought of as a virtual manipulation whose reachability is the set of allowable poses. Though the sampling for TSR Chains follows clearly from that of TSRs, the distance metric for TSR Chains is radically different. TSR Chains allow high degree-of-freedom systems such as humanoid robots to perform constrained whole-body manipulation tasks while exploiting the redundancy of these systems.














    Planning with Pose Uncertainty

    A common assumption when planning for robotic manipulation tasks is that the robot has perfect knowledge of the geometry and pose of objects in the environment. For a robot operating in a home environment it may be reasonable to have geometric models of the objects the robot manipulates frequently and/or the robot's work area. However, these objects and the robot frequently move around the environment, introducing uncertainty into the pose of the objects relative to the robot. Laser-scanners, cameras, and sonar sensors can all be used to help resolve the poses of objects in the environment, but these sensors are never perfect and usually localize the objects to be within some hypothetical probability distribution of pose estimates. If this set of probable poses is large, planning with the best hypothesis alone can be unreliable because the robot may fail to complete the task and unsafe because the robot could collide with a poorly-localized object.

    Since we would like to compute a plan that is guaranteed to meet task specifications for all hypotheses of object pose, we must modify the TSRs of a given task to account for pose uncertainty and introduce virtual obstacles into the simulation to avoid potential collisions. We start by duplicating the TSRs for each pose hypothesis and computing the intersection of all duplicates. We can then sample from the 6-dimensional polytope of intersection to obtain poses that are guaranteed to obey task constraints despite pose uncertainty. A key advantage of this approach is that if the pose uncertainty is too great to accomplish a certain task, we can quickly reject that task without invoking a planner. This approach can also eliminate grasping strategies which are not robust to uncertainty.


    Motion Planning Theory

    We were the first to present a proof for the probabilistic completeness of RRT-based algorithms when planning with constraints on end-effector pose. If the manifold of valid configurations corresponding to the pose constraint has non-zero measure in the C-space, it is straightforward to show that algorithms like CBiRRT are probabilistically complete. This is because random sampling in the C-space will eventually place samples inside of the manifold. 

    However, pose constraints can also induce lower-dimensional constraint manifolds in the configuration space, making rejection sampling techniques infeasible. Sampling-based algorithms such as CBiRRT can overcome this problem by using the sample-project method: sampling coupled with a projection operator to move configuration space samples onto the constraint manifold. Before our work, it was not known whether the sample-project method produces adequate coverage of the constraint manifold to guarantee probabilistic completeness. Our proof guarantees probabilistic completeness for a class of RRT-based  algorithms given an appropriate projection operator.  This proof is valid for constraint manifolds of any fixed dimensionality.


    Base Placement for Mobile Manipulators

    Path planning for a mobile manipulator involves multiple levels of planning which are often divided into sub-problems to manage complexity. For a pick-and place operation the break-down might be: 1) move the robot to a configuration near the object, 2) grasp the object, 3) move the robot (holding the object) to some configuration which places the object into its goal configuration. Breaking the problem into the above sub-problems reduces the complexity of the overall task by allowing sub-plans to be generated in series. However, ignoring the coupling between sub-problems can turn feasible problems into infeasible ones and introduce unnecessary difficulty for the path planning algorithm.  We show how to couple sub-problems through choosing optimal grasps and base placements for transition configurations, i.e. configurations where the robot first grasps or releases the object. We find optimal configurations (in terms of grasp quality, manipulability, and clutter) by searching a constrained space of base placements and grasps using a co-evolutionary algorithm. This algorithm significantly out-performs random sampling and is able to generate high-quality configurations in extremely cluttered environments.






  • Constrained Manipulation Planning
    Dmitry Berenson
    Ph.D. dissertation, Robotics Institute, School of Computer Science, Carnegie Mellon University, June, 2011.
    Details | PDF

  • Grasping in Cluttered Environments
    The goal of a grasping algorithm is to find a pose and configuration of a robot's hand which grasps a given object in a desirable way. Our research in grasping has focused on extracting information about an object's environment to make decisions about how that object should be grasped.
    Expand

    The goal of a grasping algorithm is to find a pose and configuration of a  robot's hand which grasps a given object in a desirable way. Development of grasp quality metrics, such as force-closure, has been intensely studied in robotics for many years; however these metrics neglect a key component of grasp selection: an object's environment. Our research in grasping has focused on extracting information about an object's environment to make decisions about how that object should be grasped. Considering the object's environment allows the robot to quickly eliminate hand poses that place the hand in areas where there are obstacles and to focus on areas where the hand is collision-free and where the fingers make good contact with the object. Our grasping algorithms are able to find grasps in environments where state-of-the-art algorithms cannot, while requiring only several seconds of computation. We have also extended these algorithms to consider two-handed grasps with similar results.







  • Grasp Synthesis in Cluttered Environments for Dexterous Hands
    Dmitry Berenson and Siddhartha Srinivasa
    IEEE-RAS International Conference on Humanoid Robots (Humanoids), December, 2008
    Details | PDF | Video

  • Grasp Planning in Complex Scenes
    Dmitry Berenson, Rosen Diankov, Koichi Nishiwaki, Satoshi Kagami, and James Kuffner
    IEEE-RAS International Conference on Humanoid Robots (Humanoids), December, 2007
    Details | PDF | Video



  • Faculty


    Assistant Professor
    Computer Science and Robotics Engineering
    dberenson@cs.wpi.edu


    Post Docs


    Post-Doc
    jmainprice@wpi.edu


    PhD Students


    Calder Phillips-Grafflin
    PhD Student, Robotics Engineering
    cnphillipsgraffl@wpi.edu
    Artem Gritsenko
    PhD Student, Computer Science
    avgritsenko@wpi.edu
    Ruikun Luo
    PhD Student, Robotics Engineering
    rluo@wpi.edu
    Antonio Umali
    PhD Student, Computer Science
    avumali@wpi.edu


    Undergraduate Students


    Alex Henning
    Undergraduate
    adhenning@wpi.edu
    Rafi Hayne
    Undergraduate
    rhhayne@wpi.edu
    John Morrow
    Undergraduate
    jfmorrow@wpi.edu
    Evan Richard
    Undergraduate
    ejrichard@wpi.edu


    Alumni

    Name
    Degree
    After Graduation
    Jessica Gwozdz
    B.S. RBE/CS 2014
    Foliage, Inc.
    Quan Peng
    B.S. RBE 2014
    M.S. at UC Berkeley
    Nick Morin
    B.S. RBE/CS 2014
    Wayfair
    Ransom Mowris
    B.S. RBE/PW 2014
    HighRes Biosolutions

      Refereed Journal Papers

    • Toward a user-guided manipulation framework for high-DOF robots with limited communication
      Calder Phillips-Grafflin, Nicholas Alunni, Halit Bener Suay, Jim Mainprice, Daniel Lofaro, Dmitry Berenson, Sonia Chernova, Robert W. Lindeman, and Paul Oh
      Journal of Intelligent Service Robotics, Vol. 7, No. 3, pp. 121–131, July 2014
      PDF

    • Robot-Guided Open-Loop Insertion of Skew-Line Needle Arrangements for High Dose Rate Brachytherapy
      Animesh Garg, Timmy Siauw, Dmitry Berenson, J. Adam M. Cunha, I-Chow Hsu, Jean Pouliot, Dan Stoianovici, and Ken Goldberg
      IEEE Transactions on Automation Science and Engineering (T-ASE), Vol. 10, No. 4, October 2013
      PDF

    • NPIP: A Skew Line Needle Configuration Optimization System for HDR Brachytherapy
      Timmy Siauw, Adam Cunha, Dmitry Berenson, Alper Atamturk, I-Chow Hsu, Ken Goldberg, and Jean Pouliot
      Medical Physics, Volume 39, Number 7, pp. 4339 - 4346, July, 2012.
      Details | PDF

    • HERB 2.0: A Personal Assistant in the Home
      Siddhartha S. Srinivasa, Dmitry Berenson, Maya Cakmakz, Alvaro Collet, Mehmet R. Dogar, Anca D. Dragan, Ross A. Knepper, Tim Niemueller, Kyle Strabala, Mike Vande Weghe, Julius Ziegler
      Proceedings of the IEEE, Vol. 100, No. 8, pp. 1-19, July, 2012.
      Details | PDF

    • Task Space Regions: A Framework for Pose-Constrained Manipulation Planning
      Dmitry Berenson, Siddhartha Srinivasa, and James Kuffner
      International Journal of Robotics Research (IJRR), Volume 30, Number 12, pp. 1435 - 1460, October, 2011.
      Details | PDF

    • HERB: A Home Exploring Robotic Butler
      Siddhartha Srinivasa, David Ferguson, Casey Helfrich, Dmitry Berenson, Alvaro Collet Romea, Rosen Diankov, Garratt Gallagher, Geoffrey Hollinger, James Kuffner, and J Michael Vandeweghe
      Autonomous Robots, Vol. 28, No. 1, pp. 5-20, January, 2010.
      Details | PDF

    • Refereed Conference Papers

    • From Autonomy to Cooperative Traded Control of Humanoid Manipulation Tasks with Unreliable Communication: System Design and Lessons Learned
      Jim Mainprice, Calder Phillips-Grafflin, Halit Bener Suay, Nicholas Alunni, Daniel Lofaro, Dmitry Berenson, Sonia Chernova, Robert W. Lindeman, and Paul Oh
      IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), November, 2014
      PDF | Video

    • A Representation Of Deformable Objects For Motion Planning With No Physical Simulation
      Calder Phillips-Grafflin and Dmitry Berenson
      IEEE International Conference on Robotics and Automation (ICRA), May, 2014.
      PDF | Video

    • DARPA Robotics Challenge: Towards a User-Guided Manipulation Framework for High-DOF Robots
      Nicholas Alunni, Halit Bener Suay, Calder Phillips-Grafflin, Jim Mainprice, Dmitry Berenson, Sonia Chernova, Robert W. Lindeman, Daniel Lofaro, and Paul Oh
      IEEE International Conference on Robotics and Automation (ICRA) Video Proceedings, May, 2014.
      PDF | Video

    • Human-Robot Collaborative Manipulation Planning Using Early Prediction of Human Motion
      Jim Mainprice and Dmitry Berenson
      IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), November, 2013
      PDF | Video

    • Manipulation of Deformable Objects Without Modeling and Simulating Deformation
      Dmitry Berenson
      IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), November, 2013
      PDF | Video

    • Toward A User-Guided Manipulation Framework for High-DOF Robots with Limited Communication
      Nicholas Alunni, Calder Phillips-Grafflin, Halit Bener Suay, Daniel Lofaro, Dmitry Berenson, Sonia Chernova, Robert W. Lindeman, Paul Oh
      IEEE International Conference on Technologies for Practical Robot Applications (TePRA), April, 2013
      PDF | Video

    • Initial Experiments toward Automated Robotic Implantation of Skew-Line Needle Arrangements for HDR Brachytherapy
      Animesh Garg, Timmy Siauw, Dmitry Berenson, Adam Cunha, I-Chow Hsu, Jean Pouliot, Dan Stoianovici, Ken Goldberg
      IEEE International Conference on Automation Science and Engineering (CASE), August, 2012. Best Application Paper Award
      PDF

    • Estimating Part Tolerance Bounds Based on Adaptive Cloud-Based Grasp Planning with Slip
      Ben Kehoe, Dmitry Berenson, and Ken Goldberg
      IEEE International Conference on Automation Science and Engineering (CASE), August, 2012.
      PDF

    • A Constraint-Aware Motion Planning Algorithm for Robotic Folding of Clothes
      Karthik Lakshmanan, Apoorva Sachdev, Ziang Xie, Dmitry Berenson, Ken Goldberg, and Pieter Abbeel
      International Symposium on Experiment Robotics (ISER), June, 2012.
      PDF | Video

    • A Robot Path Planning Framework that Learns from Experience
      Dmitry Berenson, Pieter Abbeel, and Ken Goldberg
      IEEE International Conference on Robotics and Automation (ICRA), May, 2012.
      Details | PDF | Video

    • Constellation - An Algorithm for Finding Robot Configurations that Satisfy Multiple Constraints
      Peter Kaiser, Dmitry Berenson, Nikolaus Vahrenkamp, Tamim Asfour, Rudiger Dillmann, Siddhartha Srinivasa
      IEEE International Conference on Robotics and Automation (ICRA), May, 2012.
      PDF

    • Toward Cloud-Based Grasping with Uncertainty in Shape: Estimating Lower Bounds on Achieving Force Closure with Zero-Slip Push Grasps
      Ben Kehoe, Dmitry Berenson, and Ken Goldberg
      IEEE International Conference on Robotics and Automation (ICRA), May, 2012.
      PDF

    • Addressing Cost-Space Chasms for Manipulation Planning
      Dmitry Berenson, Thierry Simeon, Siddhartha Srinivasa
      IEEE International Conference on Robotics and Automation (ICRA), May, 2011.
      Details | PDF | Video

    • People Helping Robots Helping People: Crowdsourcing for Grasping Novel Objects
      Alexander Sorokin, Dmitry Berenson, Siddhartha Srinivasa, and Martial Hebert
      IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), October, 2010
      Details | PDF | Video

    • Probabilistically Complete Planning with End-Effector Pose Constraints
      Dmitry Berenson and Siddhartha Srinivasa
      IEEE International Conference on Robotics and Automation (ICRA), May, 2010
      Details | PDF

    • Pose-Constrained Whole-Body Planning using Task Space Region Chains
      Dmitry Berenson, Joel Chestnutt, Siddhartha Srinivasa, James Kuffner, and Satoshi Kagami
      IEEE-RAS International Conference on Humanoid Robots (Humanoids), December, 2009
      Details | PDF | Video

    • Addressing Pose Uncertainty in Manipulation Planning Using Task Space Regions
      Dmitry Berenson, Siddhartha Srinivasa, and James Kuffner
      IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), October, 2009
      Details | PDF

    • Humanoid Motion Planning for Dual-Arm Manipulation and Re-Grasping Tasks
      Nikolaus Vahrenkamp, Dmitry Berenson, Tamim Asfour, James Kuffner, and Rudiger Dillmann
      IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), October, 2009
      Details | PDF

    • Manipulation Planning on Constraint Manifolds
      Dmitry Berenson, Siddhartha Srinivasa, David Ferguson, and James Kuffner
      IEEE International Conference on Robotics and Automation (ICRA), May, 2009
      Details | PDF | Video

    • Manipulation Planning with Workspace Goal Regions
      Dmitry Berenson, Siddhartha Srinivasa, David Ferguson, Alvaro Collet Romea, and James Kuffner
      IEEE International Conference on Robotics and Automation (ICRA), May, 2009
      Details | PDF

    • Object Recognition and Full Pose Registration from a Single Image for Robotic Manipulation
      Alvaro Collet Romea, Dmitry Berenson, Siddhartha Srinivasa, and David Ferguson
      IEEE International Conference on Robotics and Automation (ICRA), May, 2009
      Details | PDF

    • Grasp Synthesis in Cluttered Environments for Dexterous Hands
      Dmitry Berenson and Siddhartha Srinivasa
      IEEE-RAS International Conference on Humanoid Robots (Humanoids), December, 2008
      Details | PDF | Video

    • The Robotic Busboy: Steps Towards Developing a Mobile Robotic Home Assistant
      Siddhartha Srinivasa, David Ferguson, J Michael Vandeweghe, Rosen Diankov, Dmitry Berenson, Casey Helfrich, and Hauke Strasdat
      International Conference on Intelligent Autonomous Systems (IAS), July, 2008
      Details | PDF

    • An Optimization Approach to Planning for Mobile Manipulation
      Dmitry Berenson, Howie Choset, and James Kuffner
      IEEE International Conference on Robotics and Automation (ICRA), May, 2008
      Details | PDF

    • Grasp Planning in Complex Scenes
      Dmitry Berenson, Rosen Diankov, Koichi Nishiwaki, Satoshi Kagami, and James Kuffner
      IEEE-RAS International Conference on Humanoid Robots (Humanoids), December, 2007
      Details | PDF | Video

    • Hardware Evolution of Analog Circuits for In-situ Robotic Fault-Recovery
      Dmitry Berenson, Nicholas Esteves, and Hod Lipson
      NASA/DoD Conference on Evolvable Hardware, June, 2005
      Details | PDF | Video

    • Refereed Workshop Papers

    • Using Task Symmetry for Human-Robot Collaborative Manipulation of Deformable Objects Without Modeling Deformation
      Dmitry Berenson
      EEE/RSJ International Conference on Intelligent Robots and Systems (IROS) Workshop on Cognitive Surgical Robotics: From Virtual Fixtures to Advanced Cooperative Control, November, 2013. Best Poster Award
      PDF

    • Grasp Synthesis in Cluttered Environments for Dexterous Hands
      Dmitry Berenson and Siddhartha Srinivasa
      Robotics Science and Systems (RSS) Workshop on Robot Manipulation: Intelligence in Human Environments, June, 2008
      Details | PDF

    • Theses

    • Constrained Manipulation Planning
      Dmitry Berenson
      Ph.D. dissertation, Robotics Institute, School of Computer Science, Carnegie Mellon University, June, 2011.
      Details | PDF



    Visit us

    We are happy to arrange visits to the lab for K-12 student groups during the school year or summer. Please contact Prof. Dmitry Berenson (dberenson [at] cs.wpi.edu) to arrange a visit.

    Visit the ARC Lab on facebook!

    Visit our YouTube channel for the latest videos.



    Where is the ARC lab?

    We are located in the basement of Atwater Kent Laboratories (AK 013) at WPI. Please see this map for the location of the building.

         


    Visits to schools

    For schools in the Boston/Worcester area, we can set up a visit to your school to give a presentation on computer science and robotics. Arrange a visit by contacting Prof. Dmitry Berenson (dberenson [at] cs.wpi.edu).



    Photos from recent outreach events

    Datalink Toolkit
    A ROS package designed for remote operation of a robot over a high-latency and low-bandwidth datalink
    Expand

    The Datalink Toolkit is a ROS package designed for remote operation of a robot over a high-latency and low-bandwidth datalink. The package allows the user to easily set up relays and compression methods for a single-master system. These relays avoid duplicating data sent over the datalink while compressing common datatypes (i.e. point-clouds and images) to minimize bandwidth usage. The package was developed and extensively tested as part of the DARPA robotics challenge, though it is not specific to a type of robot.

    The package allows the user to easily set up relays and compression methods for a single-master system. These relays avoid duplicating data sent over the datalink while compressing common datatypes (i.e. point-clouds and images) to minimize bandwidth usage.

    The toolkit includes both message-based and service-based relays so that data can be sent on-demand or at a specified frequency. The service-based relays are more robust in low-bandwidth conditions, guaranteeing the synchronization of camera images and camera info messages, and allow more reconfiguration while running.

    The key features of the package are:

    • Generic relays with integrated rate throttling for all message types
    • Dedicated relays with rate throttling for images and pointclouds
    • Generic service-based relays with integrated rate throttling for all message types
    • Dedicated service-based relays with integrated rate throttling for images and pointclouds
    • Image resizing and compression using methods from OpenCV and image_transport
    • Pointcloud voxel filtering and compression using methods from PCL, Zlib, and other algorithms. (Note: pointcloud compression is provided in a separate library that can be easily integrated with other projects)
    • Launch files for easy use of the datalink software with RGBD cameras
    • Works with ROS Hydro

    For more information, please see the wiki.

    Get the package from our git repository.

    Constrained Manipulation Planning Suite (CoMPS)
    The Constrained Manipulation Planning Suite (CoMPS) consists of three openrave plugins and associated data files. The planning and inverse kinematics algorithms in this suite are designed for articulated robots like robotic arms and humanoids.
    Expand
    The Constrained Manipulation Planning Suite (CoMPS) consists of three openrave plugins and associated data files. The planning and inverse kinematics algorithms in this suite are designed for articulated robots like robotic arms and humanoids. 

    CoMPS is implemented in C++ and compiles in linux only. There are also several examples in python and matlab that show how to interface with openrave to use the plugins in CoMPS. 

    This package is available on SourceForge.
    LightningROS
    LightningROS is a ROS package implementing the Lightning Path Planning Framework. This approach uses a path library to store previous experience while allowing generality by also planning from scratch. Please see this paper for more details. This package uses OMPL planners to implem
    Expand

    LightningROS is a ROS package implementing the Lightning Path Planning Framework. This approach uses a path library to store previous experience while allowing generality by also planning from scratch. Please see this paper for more details.

    This package uses OMPL planners to implement each component in Lightning and can be called the same way as any other OMPL planner.

    This package is available on SourceForge.

    ARC Lab Github Page
    The latest, most bleeding-edge software developed in the lab.
    Expand

    Visit our Github page to get the latest code developed in the lab. Watch out, it's a jungle in there!

    © 2012 Dmitry Berenson. All Rights Reserved.