Linhao Bai

Austin, Texas

• Work across production robotics systems spanning multi-robot orchestration, distributed / edge services, robot runtime, simulation / virtual commissioning, localization, robot interfaces, and agentic AI tooling for factory automation.
• Build software-defined orchestration and integration systems for AMR, ASRS, and conveyance, translating manufacturing processes into reusable job / task abstractions, robot behaviors, configuration, and scalable engineering infrastructure.
• Work across fleet software, robot / autonomy layers, industrial control interfaces, infrastructure, and physical processes to connect system architecture with production-scale robotic behavior.

Atlanta, Georgia

• Developed dexterous robot-learning systems spanning visuomotor dynamics, behavior prediction / replanning, residual RL, simulation, and sim-to-real; contributed to K-UBM and Video2Sim2Real.
• Built a ROS-based real-time policy-execution stack for a Kinova arm with PSYONIC / Allegro hands, including low-latency joint streaming, joint-space velocity planning, synchronized arm-hand control, feedback, and evaluation logging.

Shanghai, China · Global deployments

• Delivered AMR software and integration solutions to 35+ international clients across North America, Europe, and Asia, working across fleet systems, localization, networking, embedded interfaces, and robot hardware.
• Developed C++ services for OTA / backup, time synchronization, health monitoring, integration plugins, and telemetry; worked across robot controllers, LiDAR / localization, embedded communications, and fleet software.

Going with the Flow: Koopman Behavioral Models as Pseudo Planners for Visuo-Motor Dexterity

CoRL 2026 — Y. Han, J. Fu*, L. Bai*, Z. Xiao*, et al. (*equal contribution) — Project · arXiv

Developed visual-flow representations and Koopman latent dynamics for long-horizon visuomotor behavior generation and event-triggered online replanning in dexterous manipulation.

Video2Sim2Real: Full-Stack Autonomous Dexterous Skill Acquisition from a Single Human Video

Preprint 2026 — Y. Han*, J. Qiu*, L. Bai, Z. Xiao, et al. — Project · arXiv · Code

Contributed to foundation-model perception, scene / motion reconstruction, residual RL, object-flow objectives, and real-robot policy execution in a full-stack video-to-simulation-to-real pipeline.

Additional publication: “A 1/R Law for Kurtosis Contrast in Balanced Mixtures,” IEEE Signal Processing Letters, 2026.

Built the software bridge from learned policies to physical manipulation hardware, integrating ROS control, Kinova command interfaces, dexterous hands, state feedback, timing, safety limits, and structured experiment logging.

Led a five-AMR / 100+ rack system; redesigned routing and fleet workflows, introduced hybrid SLAM + QR localization correction and rack reservation / queueing, and ran randomized multi-robot stress tests across several hundred rack movements.