Event
Kamino: GPU-parallel simulation solver supports closed kinematic loops
1 reports · 1 sources
Tsounis et al. propose Kamino, a GPU physics solver for strongly coupled heterogeneous mechanical systems, implemented in Python based on NVIDIA Warp and integrated into the Newton framework, with an ADMM-based constrained optimization algorithm at its core. The research team demonstrated reinforcement learning policy training on a bipedal robot containing six nested closed kinematic loops, simulating 4,096 environments in parallel on a single GPU and producing a feasible walking policy.
Report timeline
- Kamino: A GPU-Based Rigid Multi-Body Physics Simulation Solver with Kinematic Loops
The research team presents Kamino, a GPU-based physics solver implemented in Python using NVIDIA Warp and integrated into the Newton framework, designed for large-scale parallel simulation of heterogeneous robotic systems with strongly coupled kinematic and dynamic constraints, such as kinematic loops.
arXiv cs.RO ·