Panel Discussion: Robotics | AI Frontier: Data, Agents & Robots | SF Tech Week 2025

In this talk

1. 11:12

   Robotics goals and field applications

   Steve Maguire describes using robotics in agriculture and ecology, including planning wireless sensor placement across farm fields and accounting for the sensors’ full life cycle. Tal Chen says Dexmate is building reusable technology to make robotics applications easier to create. Leo Ma says RoboForce aims to automate dangerous, repetitive work, including jobs at large solar sites.

2. 15:46

   Data and deployment bottlenecks

   Maguire says agricultural AI needs labeled data that covers natural variation, but collecting it is costly and some disease cases cannot be ethically created for research. Ma says RoboForce has moved past some model and system hurdles, while customer adoption and supply-chain capacity remain constraints. He cites more than 35 early customer partners across 12 countries and orders for over 11,000 systems, alongside a hardware design with 15 subsystems and 660 components.

3. 22:22

   AI workflows and generalization

   Maguire sees language models as useful research and coding assistants, while stressing that students need to learn when their output can be trusted. Chen describes training robotics models with large datasets from teleoperation, simulation, and human demonstrations, then adapting them to specific tasks and hardware. The panel sees foundation models as a way to reduce customer-specific development, but Ma emphasizes that customers need reliable performance more than broad generalization alone.

4. 35:33

   Choosing robot forms and designing for people

   The panel argues that robot shape should match the job: wheels, humanoid bodies, and other designs each suit different settings. Chen says a limited range of hardware designs can help control cost, while common software and AI layers can support multiple forms. Panelists also note that humanlike appearance may help acceptance in care settings, and that robots can signal intended movement to make nearby people feel safer.

5. 43:50

   Safety, simulation, and practical AI

   Panelists describe safety as a combination of predictable motion, force sensing, collision monitoring, and safe training data. Ma says RoboForce aims to keep people out of hazardous work sites such as remote solar projects and data-center construction areas. On simulation, Maguire favors physical models when the underlying science is understood, while Chen says teams should choose simulation or real-world data according to the task and its cost.