Introductions to the open-source projects behind this site: what they solve, how they work and their limits.
Robot Reel’s agent director checks that a storyboard covers a recording with no gaps, then replays it in Blender: 4 shots, 180 source samples in 210 frames.
evalarc hillclimb-run loops your evaluate and propose commands; in a scripted example it rolls back pasted case answers and an overfitted edit.
pai-design-workbench mounts only an accepted, digest-checked CAD part on a MuJoCo arm, requires mesh mass within 3% of B-Rep, then exports MJCF and OpenUSD.
pai-design-workbench runs each OpenFOAM v2512 case as a Batch job, verifies hashes and version, then four checks. Candidate Cd: 0.22945 local, 0.22926 Batch.
Run Robot Reel’s first-use trial on your own LeRobot episode: install, export, verify, time it, report. How reports are counted; the pinned record counts 0.
EvalArc's diff, eval-health and hillclimb-review save input copies and all options; evalarc verify recomputes offline. What exit codes 0, 1, 2 mean.
Turns sim runs into browser replays, hash-verified raw data and editable Blender/OpenUSD scenes; core: a paired stress test of 30 closed-loop SmolVLA runs.
Up to 8 embodied AI items a day from arXiv, vendor blogs and media, in eight lanes, tagged O/R/M, numbers quoted as published; open rules, reproducible offline.
Compares two agent eval runs check by check under explicit acceptance rules to catch regressions hidden by a higher average, and audits graders on known bugs.
Merges Agent Skills from Claude Code, Codex, Cursor, plugin marketplaces and Git into one catalog, served via DeepSeek Harness or MCP, with file/hash checks.
Open-source WordPress plugin for Amazon Bedrock on your own AWS account: streaming chat, a controlled agent, MCP server and client, rule-based answer checks.
Sign in to chat with the assistant.
Sign in