Undergraduate Research Assistant
Published:
- Combined a novel sampling strategy inspired by Expansive Space trees with Model Predictive Path Integral (MPPI) to improve its exploration ability, increasing the target-reaching success rate in cluttered environments.
- Integrated an online-learned value field inspired by Real-Time Adaptive A* into MPPI to improve its ability to escape local minima with a limited horizon.
- Implemented and validated the algorithm with several motion-planning tasks using a UR5e robot arm and an ant quadruped robot.
- Benchmarked the implementation with CPU MuJoCo, GPU-accelerated MuJoCo MJX and Mujoco Warp.