Actuate 26 | The Full-Stack Approach to General-Purpose Robots by Theophile Gervet
Theophile Gervet, co-founder of Genesis, explains at Actuate 26 how the company combines human manipulation data, simulation, AI, hardware, and deployment to build general-purpose robots.
In this talk
- 0:11
A full-stack strategy for robotics
Gervet introduces Eno and says Genesis is pursuing robotics as a systems problem. Its approach combines scalable manipulation data, simulation, a foundation model, and a robot, with progress across all of those areas rather than just one.
- 3:14
Scale human data for manipulation
Gervet sees manipulation as robotics’ central challenge and human hand activity as the key source of training data. Genesis aims to pretrain on first-person recordings, use a few hours of glove demonstrations to adapt the model to specific tasks, and scale post-training through trial and error in simulation.
- 5:25
Build AI, hardware, and distribution together
Genesis treats AI, hardware, and distribution as connected parts of the system. Gervet says human-like dexterous hands can help transfer human data to robots, while deploying more robots can generate data and reduce manufacturing costs. The company’s Genie model is shown doing tasks such as cooking an omelet, cable harnessing, and pipe fitting; Gervet says it is trained on hundreds of thousands of human-data hours and adapted to each task with about 20 hours.
- 11:29
Use simulation to speed evaluation
Gervet argues that testing robots in the physical world is slow, costly, and difficult to reproduce. Genesis builds a simulation stack with a renderer called Nix, the Genesis World physics engine, and the Quadrants cross-platform container. It simulates rigid and deformable materials and sensors, and is designed for real-time evaluation and large-scale data generation; Genesis is open source and supports NVIDIA, AMD, ARM, and Mac Metal.
- 15:52
Design for trust and real deployment
Gervet says robot design should prioritize useful manipulation and people’s trust. Genesis emphasizes human-function arms and hands, while noting that some applications may suit wheels better than legs; Eno is designed to fold away when not needed. The company is focused on industrial deployment and has announced a partnership with LG.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.