Ben Lorica on world models: if AI builds the robot instead of controlling it, what's left?
bigdata · x · 2026-10-30
In a Gradient Flow essay, Ben Lorica revises his view on world models. Many robotics tasks fall into "variational automation" — fixed robots and workspaces, but packages, groceries, and cables vary every time — and remain expensive to automate because each application needs engineers to assemble perception, grasping, motion planning, and recovery systems, while building faithful simulators is itself labor-intensive. He contrasts traditional component-based robotics with emerging VLA foundation models, then poses a pointed question: if coding agents can assemble robot programs from reusable skills and even build the simulators to test them, what does a learned world model add that a foundation model, existing robotics software, and a physics engine can't provide?
More from Embodied
- RT-SAFE Benchmark Shows Faster Reasoning Doesn't Mean Safer Agents in Real-Time Embodied Environments — Lianhuiq · 2026-10-30
- LeCun team's LeHami: symplectic Hamiltonian flow keeps JEPA world models from collapsing latent states — ylecun · 2026-10-30
- YC-backed Aviern is building unmanned cargo planes to autonomously fly freight to all 5,000 US public-use airports — ycombinator · 2026-10-30
- Wayve CEO: the path to general-purpose robotics runs through self-driving — alexgkendall · 2026-10-30
- Neuralink patient speaks again via thought alone, with Grok Voice restoring his natural voice — TinfoilTricorn · 2026-10-30
- Foundry Robotics raises $100M Series A, opens 130,000 sq ft California factory — ashajadeja · 2026-10-30