AXIS-Bench
A benchmark platform for quality-aware VLA pretraining research, built from crowdsourced simulated demonstrations. It supports quality label recomputation, trajectory replay and customization of real-world target tasks, with the aim of providing a reproducible controlled experimental environment.
Timeline (last 90 days)
The paper constructs AXIS-Bench, based on crowdsourced simulated demonstrations, providing recorded object states and contact states, with recomputable quality labels and replayable trajectories.
The project page has been made public at https://axis-bench.github.io/, but the specific availability of code and benchmark data needs further confirmation.
All Facts (2)
The project page has been made public at https://axis-bench.github.io/, but the specific availability of code and benchmark data needs further confirmation.
50%UnverifiedThe paper constructs AXIS-Bench, based on crowdsourced simulated demonstrations, providing recorded object states and contact states, with recomputable quality labels and replayable trajectories.
50%