Physical AI News (Japan)

Physical AI, robotics and compute — reported from Japan.

日本語
· Product launches· 日本語で読む

RoboParty outlines three ways to get started with RP1, from simulation to a physical robot

RoboParty’s RP1 quick-start document organizes setup into three paths. Users can start with simulation, walking-policy training or operating a physical robot. The common workflow is model, train, export, validate with Sim2Sim and deploy on a physical robot.

What happened

RoboParty’s RP1 quick-start document presents three ways to begin: simulation, walking-policy training and a first operation of the physical robot. Users can choose a path independently, with each connecting to a common workflow from model to physical deployment.

Why it matters

RoboParty’s document offers a way to try RP1 without a robot or GPU, while specifying that training requires a workstation with an NVIDIA GPU and walking the physical robot requires a 24-axis configuration, a gamepad and a safety suspension rig. This gives users concrete information for deciding what equipment and validation stages they need before adopting the RP1 development workflow.

Analysis

The document’s distinguishing feature is that it does not limit RP1 setup to a single physical-robot configuration. Instead, it provides different starting points depending on the user’s available computing environment and whether they have a robot. The simulation path requires neither a GPU nor a physical robot, while the training path requires a workstation with an NVIDIA GPU, and the physical-robot path calls for a 24-axis configuration, a gamepad and a safety suspension rig. Rather than presenting accessibility as uniform, the document specifies the equipment required at each stage. Sim2Sim is included in the common workflow as a stage for validating a trained policy in MuJoCo before transferring it to a physical robot. The focus, then, is not simply on explaining how to run a robot model, but on a setup process that connects modeling, training, policy export, simulation validation and physical deployment. However, the document’s contents establish only where these steps fit in the process; they do not provide pass criteria for Sim2Sim validation or demonstrate walking performance on a physical robot. This structure is consistent with RP1’s design, which opens access to everything from mechanical design and control to simulation and training tools. RoboParty introduces RP1, an open-source humanoid robot with open hardware and software describes what is being opened, while RoboParty publishes RP1’s mechanical design and training resources on GitHub shows the continuity of the development platform. The new quick-start guide can be seen as organizing how to use those resources according to the environment users need. In relation to RoboParty documents steps from RP1 reinforcement learning to simulation validation, which covers training and validation procedures, this page serves as an overview of the full workflow. Questions to watch include how much the paths described in the document are shared across different RP1 configurations, the operating conditions for using a safety suspension rig on the physical-robot path, and how differences between simulation and the physical robot are evaluated. The procedures provided do not include policy performance figures or validation criteria. The publication of setup instructions should be considered separately from evidence of reproducible performance on a physical robot.

Sources

Translated from the Japanese original: /news/roboparty-rp1-no-dounyuu-tejun-wo-3-keiro-de-annai-shimyuree-174c43. The Japanese version is authoritative if they differ.