One of the projects currently under development focuses on object acquisition, or grasping. Reliable grasping performance can enable a wide range of applications in the short term, and serve as a foundation for more complex abilities, such as in-hand manipulation or object use. Our goal is to approach human-like reliability for object acquisition in unstructured environments, under realistic levels of sensor and actuation errors. We are investigating multiple approaches: grasping known objects based on a pre-computed knowledge database; grasping novel objects either based on their similarity with previously encountered objects, or by creating new models using shape primitives; learning the appearance subspace of graspable objects or object features in sensor readings; and modeling and reasoning about uncertainty due to sensor errors.

We are also interested in enabling robots to use tools, which requires stable grasping, precise contact positioning in the presence of sensor errors, and the ability to reason about the intended use for the grasped object. Consider, for example, a task requiring the use of a power drill: the robot must hold the tool securely, push the trigger, and apply force along the direction of the drill bit. We are studying ways of making such tasks reliable in the presence of noise, and transferable between different instances of the same class (such as drills of different makes and models).

Interaction with the environment is not limited to graspable or manipulable objects. A self-sufficient robot must be able to open doors (including, for example, kitchen cabinets or refrigerator doors), push away obstacles, press buttons, and so on. To accomplish these tasks, we are currently working on integrating manipulation with sensing and planning using modules such as fast and reliable motion planners that reason about obstacles, high-level task planners, grasp planners, etc. Applications also include whole-body control problems, such as two-arm object manipulation, or carrying objects through doors.


A Side of Data with My Robot: Three Datasets for Mobile Manipulation in Human Environments Ciocarlie, Matei., Pantofaru, Caroline., Hsiao, Kaijen., Bradski, Gary., Brook, Peter., and Dreyfuss, Ethan IEEE Robotics & Automation Magazine, Special Issue: Towards a WWW for Robots, 06/2011, Volume 18, Issue 2, (2011)  Download: ram_2011_datasets.pdf (2.63 MB)
Skill Learning and Task Outcome Prediction for Manipulation Pastor, Peter., Kalakrishnan, Mrinal., Chitta, Sachin., Theodorou, Evangelos., and Schaal, Stefan International Conference on Robotics and Automation, 05/2011, Shanghai, China, (2011)  Download: Pastor_ICRA_2011.pdf (1.96 MB)
STOMP: Stochastic Trajectory Optimization for Motion Planning Kalakrishnan, Mrinal., Chitta, Sachin., Theodorou, Evangelos., Pastor, Peter., and Schaal, Stefan International Conference on Robotics and Automation, 05/2011, Shanghai, China, (2011)  Download: Kalakrishnan_ICRA_2011.pdf (3.56 MB)
A Dataset for Grasping and Manipulation using ROS Ciocarlie, Matei., Bradski, Gary., Hsiao, Kaijen., and Brook, Peter IROS Workshop: RoboEarth - Towards a World Wide Web for Robots, 10/2010, Taipei, Taiwan, (2010)