MediumResearchPreprintLLM-specific
Lost Between Mind, Model Collaboration, and Motion: Understanding and Measuring Cross-Model Action Misalignment in Commercial Robots
- Published
- Record updated
Summary
Embodied robots chain a large language model, which interprets user intent, to a smaller action model, which turns that understanding into executable commands. The authors present TARS, a black-box framework that tests this command-reasoning-action pipeline and finds unintended physical behaviors. Across four commercial robots in 44 settings (133,400 executions), it reports widespread cross-model action misalignment and up to 94% verified action coverage.
Topics
Related items
- CriticalGHSA-9x8x-crjq-g64h: Jupyter MCP Server: Browser DNS rebinding can hijack the Jupyter backend through management routesSimilar attack · GitHub Advisory Database
- HighGHSA-24pw-rf5p-mgp7: Jupyter MCP Server: Unauthenticated management routes can hijack the Jupyter backend in standalone HTTP modeSimilar attack · GitHub Advisory Database
- HighGHSA-m932-63jc-4jqq: k8s-mcp-server: Missing authentication on default SSE/HTTP transport allows unauthenticated Kubernetes tool executionSimilar attack · GitHub Advisory Database
- HighGHSA-3mgh-93r5-r7jf: CodeWhale: exec_shell parallel-readonly classifier auto-approves fd/rg/git grep exec flags enabling silent RCESimilar attack · GitHub Advisory Database
- HighGHSA-qx49-fqc8-xw99: Model Context Protocol Python SDK: OAuth client could send credentials to an authorization server chosen by the MCP serverSimilar attack · GitHub Advisory Database