VSArena v0.6.0 launches a browser-based arena for evaluating embodied AI and VLA policies

NovaCoding · reddit · 2026-09-11

VSArena v0.6.0 is a major update to an open evaluation arena for Vision-Language-Action (VLA) and embodied AI policies, built on browser-native 3D physics — run a policy, watch the environment, and measure results without a local robotics simulator or physical robot.

Key updates:

The single canonical task (stacking three cubes in order) is intentionally simple: the author wants the evaluation loop itself reliable, reproducible, and inspectable before scaling to dozens of tasks. The long-term goal is an open arena with a public leaderboard, standardized environments, and physics-based evaluation.

Original post →

More from Embodied

Embodied channel →