Role Overview
For over 6 years, I have been a key contributor to the software stack of the Aeo humanoid robot, focusing on manipulation, motion planning, and behavior design.
Key Contributions
Behavior Design
- Extended reactive-action-processor behavior-tree mechanism with new manipulation nodes (grab, dispose, door operations), sustaining fleet-wide skill success rates > 95%.
- Owned manipulation pipelines for fetch-and-delivery, disinfection, and elevator interaction.
Motion Planning
- Enabled articulated collision objects concept leading to productization of Aeo delivery.
- Created online replanning infrastructure for whole-body motion planning.
- Introduced optimization-based planners via MoveIt and created custom MoveIt plugins for speed-ups.
Learning & Simulation
- Created online planning/execution pipeline to expose trajectory rollouts and reward signals for on-robot learning.
- Integrated cuRobo GPU planner for accelerated trajectory generation.
- Eliminated regressions on manipulation capabilities by covering every use-case with Gazebo integration tests in CI.