Event
MO-Playground Multi-Objective Reinforcement Learning Framework Released
1 reports · 1 sources
Neil Janwani and other authors released MORLAX, a GPU-native multi-objective reinforcement learning algorithm, and MO-Playground, a pip-installable GPU-accelerated multi-objective environment library. The authors state that this combination can approximate the Pareto set within minutes, achieving 26-271x speedup compared with traditional CPU methods and up to 19x speedup compared with previous GPU methods, while also learning Pareto fronts with better hypervolume. The team also implemented a custom BRUCE humanoid robot environment, learning Pareto-optimal locomotion policies in simulation for 6 practical objectives including smoothness, energy efficiency, and arm swing. The above speedup factors and results are all claims made by the authors in the paper.
Report timeline
- MO-Playground: Massively Parallel Multi-Objective Reinforcement Learning for Robotics
The authors propose MORLAX, a GPU-native multi-objective reinforcement learning algorithm, and MO-Playground, a pip-installable GPU-accelerated multi-objective environment library. They can approximate Pareto sets within minutes, achieving 26-271x speedups over traditional CPU methods and up to 19x speedups over prior GPU methods, while learning Pareto front hypervolumes that are superior. The team also implemented a custom BRUCE humanoid robot environment, learning Pareto-optimal locomotion policies in simulation across 6 practical objectives such as smoothness, energy efficiency, and arm swinging.
arXiv cs.RO ·