Can Humans Teach Robots to Think? Real-Time AI Control

LiveKit Portal is a production-grade stack for robot teleoperation, data collection, and remote inference, built on LiveKit's realtime infrastructure. In this video we show how to operate a robot you cannot physically reach, record clean time-aligned training data while you do it, and run learned policies over the network on real hardware. Sign up for LiveKit Cloud: https://cloud.livekit.io/signup?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=dxy26SAm_iU Modern robotics has shifted toward end-to-end learned policies and VLAs that consume a single bundled observation, one snapshot of what the robot sees and where its joints are. Keeping that bundle aligned is free on one machine but hard over the internet, because camera frames and joint state travel different paths and arrive at different times. Portal fuses them back together so your policy always sees a coherent observation and your recorded episodes stay aligned for training. What you'll learn in this video - Why operating a robot remotely is hard and how Portal solves it - How observation syncing reconstructs clean, time-aligned data - Running a single robot with multiple operators for human-in-the-loop data collection - Clean policy-to-human handoff, DAgger style - Live pipeline metrics including true observation-to-action latency - Why the same code works locally and remotely with only a URL change Spend your time on the robot, not the transport. 📚 Resources 📚 Tutorial for human-in-the-loop data collection: https://github.com/livekit-examples/portal-hitl Full API and configuration docs: https://github.com/livekit/portal Agent docs: https://docs.livekit.io/agents/?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=dxy26SAm_iU 🤝 Join the Community: https://community.livekit.io/?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=dxy26SAm_iU #livekit #robotics #teleoperation