In progress (Last updated: 05/21/25)
In partnership with RoMeLa (UCLA), supported by 1517 ventures
Building truly useful in-home service robots by hacking human habits.
Part 1: Habit Learning Cognitive Architecture. We all have countless habits, big and small, that keep our lives efficient. By using the latest machine learning technology, all of these habits can be picked up by your robot. The best helpers anticipate; not react.
Part 2: Ultramobile Wheelbase. Currently, most attention is focused on legged robots for inside the home. However, wheelbases provide greater efficiency, stability, and accuracy. Existing wheelbases require a large footprints and cannot negotiate common household and urban obstacles such as bumps and stairs.
I’m going to fix this and build the world’s best wheelbase platform.
Robot Visualization in MuJoCo
Update #2: April 20.
Mechanical design progress. Designed for manufacturability and scalability (up to 10kg allowance for robot arms). Cognitive architecture plan write-up added to site.
Update #1: April 15.
RL policy for simplified stair-climber successfully climbs standard stairs given movement primitives and stair dimensions. Video attached below for reference.
Methodology
Inspiration
I believe in robots. I also believe there are 3 ways in which they can be improved.
- Make them more autonomous
- Make them more personal
- Make them more physically capable
Throughout my life, I’ve tried my best to work on all of these frontiers, but would like to combine them all into one truly useful robot platform through this project.
I believe that habits are one of the fundamental forces through which we humans shape our lives. Why not give that same power to robots?
Previous Relevant Projects:
Autonomous wheelbase package delivery robot (Credits to bracket.bot)
Tested reinforcement learning for stair climbing robot given some specified movement primitives. Best performing policy shown, trained through custom PyTorch pipline including MuJoCo.
Readings
Interactive Continual Learning: . https://arxiv.org/pdf/2403.03462
LLM Integration for home service robots:
https://arxiv.org/pdf/2305.05658
Few Shot Class Incremental Learning:
https://arxiv.org/pdf/2404.02117