Newark, California, United States
• Expertise in robot learning, including imitation learning, reinforcement learning, and vision-language action (VLA) models such as pi0 and GR00T, applied to humanoid and manipulation tasks. • Strong background in classical robotics, with deep expertise in SLAM, probabilistic robotics, Bayesian filtering, global/local planning, navigation, and mapping systems. • Led development of Amazon Astro’s mapping and navigation stack, covering global mapping, planners, recovery behaviors, and SLAM integration. • Broad experience in robotic perception, sensor fusion, robotic vision, and system-level integration across research and consumer robotics. • Proficient in PyTorch, Python, C++, and ROS for robot learning, system integration, and large-scale robotics software development. • Strong collaborative engineering background, with experience leading teams, inventing novel approaches, and contributing patents in mobile robotics.
• Developed and applied robot learning techniques, including imitation learning and reinforcement learning, for humanoid control and robotic manipulation. • Worked on vision-language action (VLA) models such as pi0 and GR00T, exploring VLA-based policies for robot learning. • Applied VLA combined with reinforcement learning (RL) as post-training optimization methods to improve performance and generalization. • Designed and led the development of the full mapping and navigation stack for Amazon Astro, including: • - Global mapping and occupancy grid representation. • - Global and local planners with robust recovery behaviors. • - Integration of SLAM with mapping for long-term autonomous navigation. • Invented novel methods for navigation in quasi-static environments (e.g., doors, narrow passages). • Created approaches for collision detection, recovery, and exploration using visual and marker-based methods. • Contributed multiple patents in mobile robotics, mapping, and navigation.
- Worked on obstacle detection, obstacle mapping, surface terrain following and dynamic routing on 3D for a drone application. - Invented a gimbal LIDAR that allows a drone to point LIDAR device toward direction that the drone will fly. This device is a very light 3D LIDAR from 1D LIDAR. - Developed a vision based precise landing method which allows a drone to land in small spot. - Developed an algorithm to extract a roof wireframe and to detect roof hail damages and obstructions on a roof using deep learning. - Developed automatic calibration algorithm between a visual image and a thermal image.
- Invented a global localization method based on a vector distance transform and a particle filter on a SLAM framework for a kidnapping robot problem. -Developed a topological based room segmentation and recognition method for high level cleaning command. -Developed an reactive algorithm to escape from stuck situation and a wall following algorithm by predicting a wall line by a IR sensor. -Developed software for consumer autonomous robotic vacuums(BotVac), focusing on SLAM based on laser ranger finder, behavior based reactive navigation system and sensor fusion using accelerometer, IR, and laser ranger finder.
• Respond-R project focuses on the real-time collection, processing, distribution and visualization of information for prevention, preparedness, response and recovery from emergencies.