Kamil and Brian host a live roundtable with winners from the REVEL × LycheeAI × NVIDIA Isaac Sim Hackathon. Lukas Knak, Jan Olsson, Pascalis Trentsios, Loan BERNAT and Zakariea Sharfeddine share how they built a humanoid butler on UniTree G1 with GR00T, a physics-heavy claw machine rope sim, and a chess-playing arm trained by VLA. We dig into cloud setup on BREF H100 and H200, giant Docker images, data and domain randomization, teleop vs imitation learning, and the stubborn Sim2Real gap. Plus lessons for Hackathon 2.0 and what builders need next like one-click train compute.

In our very first episode, Brian and Kamil launch the REVEL Deep Tech Podcast and share why now is the moment for startups, researchers, and VCs to rethink how robots are trained. We frame the problem: data gaps, complex workflows, and robots that rarely practice on the tools people actually use and introduce REVEL’s approach: digital twins, a public-data object library, and a prompt-to-train platform that makes robot training stupid simple. We also recap lessons from the NVIDIA Hackathon, trade stories from the field, and set the tone for what listeners can expect each week: honest, builder-first conversations with the people shaping embodied AI.