Motion Planning Strategies & Frameworks

We develop motion planning algorithms that can be applied to any type of robot, from simple rigid bodies to complex articulated linkages. We abstract the particular motion planning problem into configuration space (C-space) where each point in C-space represents a particular configuration/placement of the robot. Invalid configurations (e.g., in-collision, high energy) become C-obstacles in this higher dimensional space. We can then plan the path of the (now point) robot in C-space and later transform it back to the actual robot.

Probabilistic roadmap (PRM) methods use randomization to construct a graph (roadmap) in C-space on which multiple quries (start/goal configurations) can be solved. C-PRM gives a framework for efficiently building and querying roadmaps. We also explore adaptive techniques to improve PRM performance for the given problem instance with our Feature-Sensitive Meta-Planner and our Single Shot motion planner.

Single Shot Planning		C-PRM: Customizable (and Adaptable) PRM		IRC: Iterative Relaxation of Constraints		Feature-Sensitive Partitioning & Planning
IMG: Incremental Map Generation		OBRRT: Obstacle-Based RRT		User-guided motion planning		RESAMPL: Region Sensitive Adaptive Motion Planner
Moving Objects		Improving Connectivity

Specialized Algorithms

We design algorithms for specialized classes of robots. These algorithms are more efficient than general purpose motion planning algorithms. For some robots, such as closed chains and foldable objects, the probability of randomly sampling key configurations is near zero. Other robots, like deformable objects, nonholonomic robots, and metamorphic robots, have unique capabilities/requirements that cannot be adequately expresssed with general purpose motion planners. Also, we work on incorporating mobile robot localization with the motion planning algorithm.

Closed Chain Systems		Deformable Objects		Foldable Objects		Nonholonomic Car-like Robots
Metamorphic Robots		Mobile Robot Navigation & Localization

Randomized Sampling

Probabilistic roadmap (PRM) methods use randomization to construct a graph (roadmap) in C-space on which multiple quries (start/goal configurations) can be solved. Sampling is one important components of PRMs. Our work on OBPRM and MAPRM provides new sampling strategies to handle more challenging narrow passage problems. RESAMPL transforms an intital sampling distribution to increase sampling in high entropy areas. We have also studied now to combine existing samplers by biasing them with each other to improve performance. Finally, we use metrics to determine how sampling methods perform and how to adapt them.

OBPRM: Obstacle-Based PRM		MAPRM: Medial Axis PRM		RESAMPL: Region Sensitive Adaptive Motion Planner		Biasing Samplers
Metrics