ManiGuard is a safety benchmark for foundation-model robot manipulation, evaluating policies on contact-rich household tasks under formal specifications across simulators and a real Franka arm.

An online generative Bayesian filtering algorithm for manufacturing state estimation that uses conditional VAEs as learned likelihoods, demonstrated on laser powder-bed fusion defect monitoring.

A two-tier embodied-agent framework in which a VLM planner orchestrates fresh-context navigation, manipulation, and verification agents across swappable simulator, policy, model, and tool backends. Verification-gated recovery and a resumable evaluation pipeline backed by 402 tests improve success over flat-policy baselines by 10% on VLN-CE navigation and 33% on BEHAVIOR-1K long-horizon mobile manipulation.

An open-source NVIDIA Isaac Lab framework for the LeRobot SO-101 arm that unifies PPO with distillation, ACT, and Diffusion Policy training. A shared joint-position action space and real-arm deployment bridge allow one checkpoint to run in simulation and on a Feetech follower without sim-to-real action conversion.

A 6-DOF hexapod with tripod and wave gaits driven by inverse kinematics and IMU-based PD feedback, paired with A* planning and a subsumption-based controller for real-time obstacle avoidance. A frontier-based exploration pipeline incrementally builds the occupancy map and prioritizes frontiers by information gain for autonomous coverage.