coding-agent-robot-manipulation-where-safety-constraint-do-touch-named CLM-0512For coding-agent robot manipulation where a safety constraint (do not touch a named obstacle) is given only as text in the task prompt, the agent restates the constraint early and then plans the shortest path through the obstacle, and enforcing the same constraint with an external stateless verifier that re-checks each route against grounded obstacle boxes before motion raises both task success and collision avoidance.
Capability: Turning plans into physical action
Observed on
Simulated tabletop manipulation (SafeLIBERO, 32 tasks, 10 seeds) with a single declared obstacle per scene, parallel-jaw gripper, frozen pi-0.5 policy, GPT-5.5/GPT-6 planners..
Sources
- Validated by: On SafeLIBERO with 10 seeds per task across 32 tasks: 71.9% task success and 87.5% collision avoidance with GPT-6, versus 31.0/59.0 for the same agent with skills but no harnesses, and 67.5/68.9 for AEGIS (barrier-function filter over the frozen policy); ablation holds prompt, skill memory and VLA fixed.. Measured on one simulated benchmark with one obstacle per scene; ablation isolates the harnesses from model scale by holding prompt, skill memory and VLA fixed. [truncated]
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- In VLM-driven human-aware robot task planning over scene graphs, the bottleneck is inferring and grounding the symbolic goal, not the planning: once both human and robot goals correctly cover the ground truth, a formal multi-agent planner produces a valid, conflict-free joint plan in 99.6% of trials, while end-to-end success with VLM-inferred human goals is 69.7%.Planning a sequence of actions before acting · reviewed by AI
- In LLM multi-agent systems that reflect after failure, restricting reflection to the single agent responsible for the first incorrect step — located by a trained step-level correctness scorer and paired with a counterfactually generated corrected step — raises task success more than having every agent reflect on the trajectory, because reflections written by agents that behaved correctly enter their memory as misleading guidance for later tasks.Getting more from several models than from one · reviewed by AI
- For LLM-driven robot manipulation where every tool call is schema-validated and every motion kinematically and collision-checked before execution, execution-time recovery mechanisms (retrying a failed plan with an alternative candidate, resetting to a collision-checked safe pose between stages) contribute more to end-to-end task success than canonical action sequencing, because most residual failures are transient planning or IK failures rather than wrong orderings.Turning plans into physical action · reviewed by AI
- For open-source LLM agents driving a ROS robot, having the model emit a pseudo-code tool-call plan up front, storing it in a persistent scratchpad, and then restricting each tool invocation to the plan's next step raises task success most on the longest multi-step tasks and for the smaller open models (GPT-OSS-20B on the hardest task rose from 35% with a persistent plan alone to 55% with the constraint), while a strong proprietary model gains little.Completing tasks that take many steps or many hours · reviewed by AI
- For multi-turn agentic coding sessions, constraints written into a repository policy file (e.g. CLAUDE.md) hold up better over successive turns than the same kinds of constraints stated in user turns, and constraints that replace an earlier constraint are followed less often than newly added ones.Following an unfamiliar procedure · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: An agent given only the textual constraint, with no external verification tool, reaching comparable collision-avoidance rates on the same benchmark, or the harness failing to improve collision avoidance over the skills-only agent under a matched step budget. Drafted stance toward fixes-in-the-harness-compound-fixes-in-the-prompt-do-not: supports -- The constraint stated in the prompt is dropped by step three hundred while the same constraint encoded as a verification tool holds across the episode.