Know why your robot failed.
Turn every camera, sensor and telemetry stream into structured context.
Curate data and root-cause failures in minutes, not days.
From raw rollouts to root cause.
RoboLens ingests every modality. Custom VLMs analyze the video, sensor analysis catches what video can't, and the agent reasons across all of it to surface failures and the data that fixes them.
Every signal, on one timeline.
Every modality, synced into one searchable record and analyzed independently. When they disagree, that disagreement is itself a signal.
Query episodes from your coding agent.
Robolens speaks MCP, so Codex, Claude Code, or any MCP client can surface failures, pull insights, and curate data, right from your episodes.
{
"mcpServers": {
"robolens": {
"command": "uv",
"args": ["--directory", "/path/to/robolens", "run", "robolens", "mcp"]
}
}
}Scriptable in your pipeline.
The same tools your agent calls are scriptable from the command line: query, discover patterns, and build datasets in CI.
# query the knowledge base
uv run robolens query --quality-min 0.5 --outcome failure
# discover cross-run failure patterns
uv run robolens patterns
# curate a training set, weighted by root cause
uv run robolens build-dataset --outcome failureRun with the teams who feel it.
We're onboarding a small number of teams deploying learned policies. If that's you, come build with us.