Designing the product that can do anything
summary
Claims a humanoid industrial robot (Atlas) becomes an economics solution by turning hardware gaps into software problems and shows the tooling that makes it work: reinforcement learning for balance and motion with 100 Hz state prediction, behavior cloning for dexterous picks and places at ~30 Hz, VR embodiment and whole‑body tracking to collect high‑quality demonstrations, and an Orbit platform for retasking and fleet monitoring. Demonstrations and deployments with Spot Stretch and Hyundai factories prove the approach scales from single skills to production orchestration in a software‑defined factory.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:10Atlas and industrial design
- 3:29The importance of the face
- 4:41Engineering for production
- 6:39Software platforms
- 9:28Training robot behaviors
- 12:06Immersive data collection
- 13:47Agentic system design
- 17:15Data scaling challenges
- 19:08The software-defined factory
related talks
21mVR and Robots: Solving the future's problems
15mAugmenting the Capabilities of Industrial Workers at Scale
32mNext generation human character control
17mRadically Evolving The Way Industries Work
12mAR for Industry 4.0: Turning Knowledge into Skill
10mAR for Optimization of Processes in Intralogistics
16mBuilding a Practical Roadmap for Success
18mReal Industry 4.0 AR Examples