Carnegie Mellon Robotics Institute
Automated design for robotics. Robot configuration design is hampered by the lack of established, well-known design rules, and designers cannot easily grasp the space of possible designs and the impact of all design variables on a robot's performance. Realistically, a human can only design and evaluate several candidate configurations, though there may be thousands of competitive designs that should be investigated. For my PhD I developed Darwin2K, an automated synthesis system for robot design. This extensible system uses an evolutionary algorithm to synthesize robots meeting a set of performance requirements for a task. Each candidate robot design is evaluated in simulation on the task at hand, and a modular software architecture allows new simulation capabilities and robot component models to be easily added.
Safeguarding and obstacle detection. My work on the Autonomous Loading System focused on the development of sensor-based safeguarding algorithms for a high-speed autonomous excavator. The challenges presented by autonomous excavation include the need for full-speed operation and reliable safeguarding, as high productivity is the main goal of the robot but accidents cannot be tolerated. The Workspace Monitor is one of two parts of the robot's safeguarding system, and continually monitors rangefinder data to detect unanticipated people and vehicles before they enter the robot's workspace. Simultaneously, the Collision Detector predicts the robot's future states based on the controller and current robot state, then checks the future states against sensory data to detect potential collisions before they occur.
Map and model registration for navigation. During a summer at JPL I worked on the Sample Return Rover (SRR), investigating two related approaches to improve the SRR's ability to rendevous with other vehicles. The first approach uses the Iterative Closest Point (ICP) algorithm to register subsequent depth maps (acquired by stereo cameras) to augment motion estimates from dead-reckoning. The second approach is also based on ICP and registers sensor data to a model of another rover with which SRR must rendezvous. The registered model pose is used to servo the SRR to a desired location relative to the other rover.
|Research Interest Keywords|
|computer vision, design, field robotics, genetic algorithms, space robotics|
|The Robotics Institute is part of the School of Computer Science, Carnegie Mellon University.|
Contact Us | Update Instructions