Modern Robot Learning from Scratch - Lecture 7 - S0-101 Presentations | Demo Day | Modern Robot Learning from Scratch

The last lecture of our robotics bootcamp was one of those sessions that reminds you why hands-on learning matters. We had 6 presentations, and every team worked with the actual SO-101 robotic arm - not simulations, not mock setups. Real hardware. Real constraints. Real learning. What made the session special was the diversity in the room. Participants came from industry, some were building their own startups, and everyone brought a different way of thinking to the table. It naturally turned into a highly collaborative discussion around what actually works when you move from theory to hardware. For the core task, we focused on a pick-and-place problem using the ACT policy. Instead of treating the baseline as fixed, teams experimented with multiple variations—tweaking policy parameters, observing failures, re-running experiments, and comparing behaviors. A lot of the learning didn’t come from success, but from why something didn’t work: (1) How sensitive the policy is to certain parameters (2) What breaks first when assumptions don’t hold (3) How small changes can dramatically affect stability and performance (4) This kind of fundamental, exploratory tinkering is hard to capture in papers - but it’s exactly what builds intuition. Seeing people argue over results, debug together, and build working applications on real robots was incredibly energizing.