Towards Unobtrusive Physical AI: Augmenting Everyday Objects with Intelligence and Robotic Moveme...
summary
System augments everyday objects with onboard motion via robotic wheel platforms and a pipeline that streams camera frames to an off‑the‑shelf VLM for captions, maintains a memory-based temporal understanding, queries an LLM to infer user goals, grounds actions with an object knowledge base, filters outputs with an action‑goal alignment check, and executes A* path planning with Bluetooth motor commands. Built prototypes move trivets, knives, key trays and push dongles; evaluation (4 actors, 4 annotators, 16 raters) matches human decisions and outperforms a VLM baseline.
AI-written summary · about
chapters (7)
- 0:00Introduction and vision
- 1:27Motivation for physical AI
- 3:13System overview and examples
- 6:20Technical architecture
- 8:29Evaluation and results
- 9:11Challenges and future work
- 10:15Q&A session
related talks
18mRobotics: why now?
03mMaking good robots
17mBeyond Automation: AI and the Evolution of Robotic Intelligence
16mReal-World Robotics as the Next Frontier for AI?
28mPreference Learning from Minimal Human Feedback for Interactive
47mDr. Simon Müller | AI Track | Compass AI & Data Summit 2025
15mPhysical AI in a World of Humans
30mThe robot in my living room - insights from a journey through the possibilities...
