Humanoid Motion/Task Planning
Here we introduce the Randomized Possibility Graph (RPG) which provides a scalable approach to utilizing probabilistically complete planners on highly constrained platforms like humanoids. The RPG works by evaluating the possibilities of various routes before performing computationally expensive whole body inverse kinematics.
This work combines task and motion planning for humanoids, using the Hybrid Backward-Forward method by Caelan Garrett.
Here we generate motion plans for lifting large/heavy objects in the presence of obstacles. We introduce a parameter called the Virtual Task Dimension which allows the robot to plan out how it can avoid obstacles while shifting its weight to lift and release heavy objects.
Miscellaneous Robot Videos
During the early stages of the DARPA Robotics Challenge, our lab had the pleasure of hosting Matt Zucker during his spring break, and we got our HUBO2+ to walk from scratch in that week:
An unscripted video of me testing a simple whole body motion planner which can generate trajectories between user-defined waypoints while avoiding collisions and remaining balanced. The planner is based on CBiRRT.