Robotics Should Belong to Everyone | Rafa Felix (runibiLABS)

In this talk

1. 0:07

   Why robotics needs a safety layer

   Felix draws on a decade of work on zero-shot generalization, including robotic vision research and safety work at TikTok. He says robot safety can benefit from lessons in digital AI, where model quality depends on the data available and how systems are evaluated. A central challenge is teaching robots to recognize dangerous interactions without exposing people to risk.

2. 2:23

   Barriers to deploying robots

   Felix describes the company’s beta robot as an early prototype and says real-world deployment still requires site preparation, calibration, and controlled conditions. He identifies three barriers for customers: high upfront hardware costs, scarce real-world training data, and low trust when people do not understand a robot or its data practices.

3. 6:35

   Web simulations as an entry point

   RunibiLABS offers a browser-based robot simulation that people can try without installing software, coding, or setting up infrastructure. Felix says the platform collects simulation interactions under data privacy practices. It uses those interactions both to help train robots on tasks and to learn how people want to interact with them.

4. 9:15

   Zero-shot learning and safer training

   Felix explains zero-shot learning with the example of a robot handling a gift it has never seen, using language, vision, and other senses to relate new objects to familiar concepts. He also describes the sim-to-real gap, where sensor differences and physical conditions can make simulation-trained behavior unreliable in the real world. His proposed approach combines generic and augmented data, simulation, and methods for learning from noisy labels, so robots can learn risky scenarios without dangerous human demonstrations.

5. 13:04

   Three pillars and ways to partner

   Felix describes a safety process in which users teach new tasks in simulation and a digital twin checks constraints before the robot performs them. He presents three pillars: research and development on sim-to-real transfer, the modular Origami robot, and enterprise partnerships that let companies test robot designs and gather user feedback. RunibiLABS is seeking pre-seed investment and commercial pilots, and invites companies to bring robots to its platform.