Field Robot Systems

Algorithms are only half of robotics. The other half is the unglamorous, essential work of building systems that survive contact with the real world: sensor drivers, state estimation, remote operation, safety layers, batteries, and weather.

Representative deployments

EnRicH 2025 — nuclear power plant inspection. With the RAICAM team, I built a low-cost aerial + ground robot system in three weeks and fielded it inside the Zwentendorf nuclear power plant at the European Robotics Hackathon: SLAM in a GNSS-denied containment structure, mesh networking through heavy shielding, and real-time radiation monitoring. Published at IEEE SSRR 2025 with a full open-source release.

Beijing 2022 Winter Olympics torch relay robots. We modified hexapod robots for skating and skiing, designed the control algorithms for both sports, and built the environment-perception and remote-control stack. The robots were tested outdoors at real skating rinks and ski resorts — an end-to-end exercise in making learned and model-based controllers reliable in freezing, unstructured conditions.

Qingzhui hexapod research platform. The capability-learning and locomotion-control research in my other themes runs on Qingzhui, which our group maintains as a complete experimental system: simulation environment, controller stack, and hardware.

Engineering philosophy

I write most systems in Python and C/C++ on ROS, simulate in MuJoCo and CoppeliaSim, and treat the simulation-to-hardware pipeline as a product in itself: if an experiment cannot be reproduced from a clean checkout, it is not finished. This theme is less about individual papers and more about the engineering standards that make the papers possible.