Interpretable Task Planner in Sim: LLM (gpt-oss) generates stepwise pick-and-place plans plus explicit intermediate traces (observations, beliefs, confidence). Sim (Isaac/ROS) executes, you log LLM traces + sensor data + video clips and run correlation/causal analysis between traces and success/fail.
Vision-Lite Robot Narrator: Edge device streams compressed structured telemetry + event tokens to gpt-oss (remote or lightweight on-device) which returns human-readable narration and structured explanations for a UI or voice output.
Safe Policy Explainer: Post-hoc explainer that consumes state/action histories and produces rationales, counterfactual “what-if” checks, and policy-safety suggestions (with simulator-backed checks where possible).
Log in or sign up for Devpost to join the conversation.