Great talk. Particularly given the last question/answer, and the work being done in the commercial space building robot simulator technology (NVIDIA, AWS, others).
I dropped out to work on motion cueing and specifically because of RL via. simulation, it's been approachable. What used to be state of the art (which in hindsight, just 5 years ago seems so 👣i.e. no sensors or feedback in the control-loop), to be influenced so easily is another testament to the quiet revolution he's talking about. The reason is the resources it took: (1 person) + Internet + LLMs eventually. Excuses are, not to use AI 🤷♂
Am going to assume that's what meant by simulation is 2-d. And am going to assume that the challenge was to move from 2-d to 3-d. If so, I would have appreciated a little insight into how this was done. There are geometries that are used for robot movement, such at cat(0) complexes, that you can analytically perform robotic movement in 2-d space but you can't in 3-d (even with sampling methods). So, just a few math equations to explain what's actually happening would have been appreciated.
5.8mil 78 likes. This is amazing an freaking. Military maby are looking for this with attention. Imagine bigger flying objects moving like that autonomously every where.
