Aakash Madabhushi

Robot Learning Engineer

I teach robot arms to do tasks on their own from human demonstrations, and build the testing that says honestly whether they work.

92%: pick-and-place success, 50 trials on a real arm at the trained cube position
2.49 cm: average distance from target on successful placements
49,633: frames of robot demonstrations published as an open dataset, across 50 episodes

Taught an SO-101 robot arm to pick up a cube and place it in a bowl on its own: 46 of 50 real trials succeeded, 2.49 cm from target on average, using a transformer policy (ACT) trained on 50 human demonstrations.
Kept a checkpoint with 67% higher action error from being used by choosing checkpoints on task accuracy rather than training loss, and ruled a larger vision-language-action model out of real-time control at 251 ms against 18 ms a decision.
Built a test harness that runs 100 trials with nobody watching: it resets the arm, recovers from crashes, and logs why each miss happened. That trace put every failure on perception, and found a motor firmware bug that reassigned device IDs.
Stated up front: the 92% is at the one cube position all 50 demonstrations used. Fifty trials at measured offsets, which will show whether the policy reads the scene or memorised a motion, are not in yet.

3 projects in this area

Every link goes to code or data you can inspect yourself.

Mar 2026 to presentLab research
Turns 50 human demonstrations into a policy a robot arm runs on its own, scored against a protocol fixed before testing.
Aug 2026 to presentPersonal project
Copies the real table in front of the robot into a simulator, so a policy can practise against the layout it will actually meet.
Mar 2026Personal project
Gets a robotics simulation stack running on a Mac or a Linux box in one command, instead of an afternoon of driver errors.

Select a project for its figures, stack and links.

I'm available from December 2026 and open to relocating. Email is the fastest way to reach me.

Hiring for something else?