Trajectory optimization shows great promise for efficiently developing feasible motion planning trajectories for robot execution. However trajectory optimization-based planning are local methods which are known to suffer in the presence of realistic clutter. Sampling methods, while guaranteed to eventually produce a suitable kinematic solution are unsuitable for use with a balancing humanoid robot as well as taking unfeasibly long to generate a solution. However can sampling of dimensional representations (e.g. the pose of only the end effector) be used to successfully seed trajectory optimization? Can a small library of common seed postures efficiently enable successful grasping in common configurations? If possible, application to cloud computing (via parallel optimization from multiple seeds) could demonstrate that parallel instances of trajectory optimization could be used to produce an end-to-end improvement in optimization time.
- Hongkai Dai, Andrés Valenzuela, and Russ Tedrake. Whole-body motion planning with centroidal dynamics and full kinematics. IEEE-RAS International Conference on Humanoid Robots, 2014.
- John Schulman, Jonathan Ho, Alex Lee, Ibrahim Awwal, Henry Bradlow and Pieter Abbeel. Finding Locally Optimal, Collision-Free Trajectories with Sequential Convex Optimization. RSS, 2014.
Research interests: Humanoid Robotics and Motion Planning