Kyle Vedder: Robot‑Learning Route
Researcher Kyle Vedder urged practitioners to break into robot learning without a PhD — buy a $300 SO101, train models, implement papers, and run on-robot evaluations — a practical, hands-on path argued. The thread pushes executable, low-cost experimentation over credential-only approaches.
Hugging Face unveiled the SO‑101 on April 28, 2025 and listed a base price of $100 while assembled or regionally‑shipped units commonly retail up to roughly $500 because of assembly premiums and tariffs. techcrunch.com The SO‑101 hardware uses a leader/follower teleoperation design and the official assembly notes specify the follower arm uses 6× STS3215 motors with a 1/345 gearing while the leader uses differently geared motors to allow low‑effort teleoperation. huggingface.co Hugging Face’s LeRobot project publishes end‑to‑end software, including SO‑101 docs and an integration into a ROS 2 workspace that bundles bridge nodes, policy nodes, teleop utilities, and dataset tooling for imitation learning. github.com Commercial kit and assembled unit distribution names cited in launch coverage include Seeed Studio, WowRobo and PartaBot as official partners for parts, kits and preassembled SO‑101 units. techcrunch.com Active sim‑to‑real pipelines and tutorials demonstrate training policies in NVIDIA Isaac / GR00T and exporting them to SO‑101 hardware, and community guides show on‑robot inference using Jetson AGX Orin for edge deployment. wiki.seeedstudio.com Kyle Vedder—researcher at Physical Intelligence and former Dyna Robotics engineer—published a December 2025 write‑up titled “State of Robot Learning” that documents the dominance of behavior cloning pipelines and the practical toolchain elements (data collection → training → on‑robot eval) he highlights as essential. vedder.io
