kevin_zakka releases mjbatch to run thousands of MuJoCo sims in parallel on CPU
KyleMorgenstein · x · 2026-09-10
Kevin Zakka open-sourced mjbatch, a Python library that steps thousands of MuJoCo simulations in parallel entirely on CPU — no GPU required.
Key features:
- C++ thread-pool execution with the GIL released; threads default to every logical core
- Live NumPy-style array access to simulation states (qpos) and controls (ctrl) across the whole batch, updated in place
- Per-simulation model parameters, with setconst to recompute derived constants (e.g., randomizing friction per env)
Usage is minimal: create Batch(model, numsims=4096), then loop ctrl[:] = policy(qpos); batch.step() to advance all 4096 sims at once. The repo ships self-contained examples showing applications from RL to system identification.
More from Embodied
- Puffin-World: unified world model with native 3D states, open-sourced with 16M dataset — ccloy · 2026-09-10
- Surgical robots matter more for China's future than humanoid robots, argues poster — PAstynome · 2026-09-10
- Medtronic Invests $700M in Chinese Surgical Robot Startup Cornerstone — PAstynome · 2026-09-10
- iPhone 18 Pro's variable aperture: Apple returns to physics after hitting compute wall — r0ck3t23 · 2026-09-10
- UISEE's AI Driver Subscriptions Grow 6x, Now 15.4% of Revenue as It Joins Stock Connect — 创业邦 · 2026-09-10
- Lirui's electrochromic film ships in Li Auto MEGA, teams with AAC on AR glasses — 创业邦 · 2026-09-10