Benchmark
MetaWorld
A widely used meta-reinforcement learning benchmark that provides a variety of robotic arm manipulation tasks and controller configurations. It is used to evaluate the generalization of methods across different control settings.
Timeline (last 90 days)
Oct 26
Backtrack was evaluated on the LIBERO and MetaWorld benchmarks, covering three VLA models: SmolVLA, π0.5, and π0-FAST.
Unverified50%
Oct 26
On MetaWorld, all twelve controller configurations tested with SmolVLA outperformed the unassisted policy.
Unverified50%