Beyond Automation: AI and the Evolution of Robotic Intelligence
summary
Generative foundation models are paired with classical control and reinforcement learning to move robots from demos into usable systems: EvoBot (double inverted pendulum) handles luggage and garbage, mixes model‑based control with RL for variable payloads, and a simulation digital twin lets policies transfer from sim to real. A pallet transporter Odin 10 m/s is reproduced in simulation. An agentic system with code generation and vision‑language monitoring demos sorting apples/quinoa/nachos. Remaining gaps: lab robustness, speed versus safety tradeoffs, and integration effort.
AI-written summary · about
chapters (14)
- 0:00AI and robotic intelligence
- 1:04Robotic foundations beyond AI
- 2:50Classical control in robotics
- 4:13Computer vision for perception
- 4:34Real-world robot environments
- 5:58Reinforcement learning for robotics
- 6:41Simulation and digital twins
- 8:04Generative AI for robotics
- 9:10Agentic systems and NLP
- 12:20Mobile manipulators for picking
- 13:22Lab demos to real-world robotics
- 14:51Speed safety in robotics
- 15:53Human acceptance in robotics
- 16:14Conclusion and research
related talks
16mReal-World Robotics as the Next Frontier for AI?
18mRobotics: why now?
16mOn the ethical challenges raised by robots powered by AI
32mBeyond the Screen: How Embodied Manifestations of AI can Amplify People in the Physical World
32mThe Road to General AI
18mOn the History and Future of AI
01h 35mDeep Robotic Learning
33mThe New Generation of Robots