Technical lead (sensor fusion, object detection and tracking, and temporal modeling)

Working on various object detection and identification approaches for millions of products in the AmazonFulfillment Center catalog (both singulated and clutter).

Deep learning and imitation learning

• Developed a navigation map system using Open Street Map. • Designed a conditional imitation learning framework for real-time GPS following driving.• The system mimicked human-like driving behaviors by following navigation commands - lane following, lane changing, turn during intersection, exiting a highway, etc.

• Designed and trained a CNN network for real-time lane following driving, running at over 20fps.• Achieved 98% autonomy in a 40-minute driving and generalization to the production environment - various road types, textures and weather conditions.

• Skeleton fitting human body shape by using the Expectation-Maximization algorithm• Point to point registration between different human poses and different identities by using non-rigid Iteration Closest Point Algorithm• Estimated the skeleton joints of a human articulated model and aligned correlated rigid parts between different pose shapes• Learned pose deformable models of human shape by encoding muscles and implemented skeleton-based body shape deformation

• Developed CUDA programming based real-time object detector using a deformable parts model.• Improved system runtime from 1fps to 10fps without compromising system accuracy

• Real-time human facial expression tracking• Dynamics texture mapping

• Designed the 3D game programs which cover the collision detection, GPU rendering and physical simulation• Grading the students' homework and projects

• Accomplished 60,000-line C# project in 2 months, developed the software platform for Evacuation Simulation• Designed the real-time evacuation optimized algorithm by using reinforcement learning algorithm

• Designed the algorithm of traffic sign recognition by using the Sparse Representation method to improve the robustness in shading or different illumination situations• Developed a probability model to estimate the position of the object, knowing the location under the World Coordinate System in the view plane

• Designed the hardware system of the whole project including the power system, mother board, the interface of network, etc.• Ported Linux Kernel to ARM and realized the communication between client and server.• Designed the program of control FPGA in VHDL