Undergraduate Research Assistant
Published:
- Designed and implemented a novel sampling strategy for Model Predictive Path Integral (MPPI), increasing target-reaching success rate by 20% in cluttered dynamic environments.
- Implemented GPU-parallel MPPI rollouts in MuJoCo Warp, achieving a 10× speedup over CPU-based simulation.
- Integrated an online-learned heuristic map, inspired by Real-Time Adaptive A*, into MPPI, reducing the required escape planning horizon from 1.1 seconds to 0.2 seconds.