Benchmark Overview

With the collected tasks, assets, and trajectories, RoboVerse establishes standardized benchmarks for robot learning, including both imitation learning and reinforcement learning. We define a unified training and evaluation protocol within the RoboVerse platform and implement standardized baselines and learning frameworks for benchmarking. Specifically, for imitation learning, we introduce different levels of generalization benchmarks to assess the generalization capability of models.

The generalization benchmark is based on randomization options that automatically applied to any task. We implemented it as an environment wrapper.