MotionForge
August 2026 — Present
GPU-accelerated multi-agent adversarial pursuit games in simulation for pretraining agile legged locomotion policies.
PythonMuJoCoWarpReinforcement LearningSelf-Play
MotionForge is the funded continuation of my CoRL workshop paper on MJTag. It scales up multi-agent adversarial pursuit games in GPU-accelerated simulation to pretrain agile legged locomotion policies through competitive self-play.
Highlights
- Builds on the MJTag environment and league-based self-play from the CoRL 2026 workshop paper.
- Supported by $5,000 in department research funding.