Cross-benchmark evaluation: DimSpec system + LIBERO/CALVIN reproduction fixes#19

Merged

MilkClouds merged 46 commits intomainfrom

cross-bench-eval-v2

Apr 1, 2026

Commits on Mar 30, 2026

feat: cross-benchmark eval configs, SimplerEnv improvements, initial results

MilkClouds
and
claude
committed
docs: add cross-benchmark pipeline verification audit

MilkClouds
and
claude
committed
feat: add DimSpec interface convention system

MilkClouds
and
claude
committed
feat: implement DimSpec declarations on all model servers and benchmarks

MilkClouds
and
claude
committed
fix: cross-benchmark reproduction fixes for X-VLA and GR00T

MilkClouds
and
claude
committed
docs: update cross-bench audit with critic review findings

MilkClouds
and
claude
committed
refactor: use obs_params for max_episode_steps instead of per-model YAMLs

MilkClouds
and
claude
committed
style: collapse single-line-eligible spec imports

MilkClouds
and
claude
committed
feat: cross-validate server↔benchmark specs via HELLO handshake

MilkClouds
and
claude
committed
refactor: code review cleanup

MilkClouds
and
claude
committed
docs: remove hardcoded local paths from audit document

MilkClouds
and
claude
committed
fix: replace numpy-unsafe `or` chain in _obs_state_array

MilkClouds
and
claude
committed
fix: SimplerEnv state — send base_pose+tcp_pose, compute base-relative EE

MilkClouds
and
claude
committed

Commits on Mar 31, 2026

Commits on Apr 1, 2026