Fully Open Humanoid Robots: Who Should Get Started?

The PM01 isn't suitable as an out-of-the-box robot; it's for teams with algorithms, engineering capabilities, and a willingness to establish safety boundaries. It's more like a high-dynamic development platform than a finished product ready to replace human labor.

Last week, I watched the PM01 demo at an exhibition and got an operation video from sales. Standing 138cm tall, weighing about 40kg, with 24 degrees of freedom (DOF) across the body and 320-degree waist rotation. These specs sound abstract, but breaking them down: DOF refers to the number of independently moving joints. 24 means its motion dimensions are more complex than many display robots, and the 320-degree waist rotation makes turning and bending more flexible.

What surprised me was how heavily it emphasizes openness. Zhongqing calls it a "fully open general-purpose embodied intelligent agent." The materials mention fully self-developed joints, dual encoders, and localized air cooling. Dual encoders mean finer joint position feedback, while localized air cooling helps dissipate heat from hot joints. For product developers, this matters more than backflips because research and industrial scenarios fear precision drift and motor overheating after long debugging sessions.

I didn't bring the robot home; my trial perspective involved watching sales videos, checking specs, and reviewing second-hand evaluation materials and developer guides. The sales video showed a complete workflow, including startup, loading actions, execution, and error prompts. The status panel roughly displays joint states, command execution, and error info—it looks more like a robot development host computer than consumer electronics where you click once and it works.

I hit three main bottlenecks. The flashier the dynamics, the harder the tasks. Running and backflipping prove control capability but don't equal stable performance in handling, inspection, or picking. Secondary development requires manpower. Interfaces, simulation, real-machine debugging, log playback—if no one on the team does this work, buying it just gets you a display unit. High-value scenarios need validation. When I worked on AI imaging at United Imaging, hospitals first asked if the model was accurate, then if it could integrate with PACS (hospital imaging systems), if reports could be written back, and who was liable for missed diagnoses. With robots, the questions become: Has the scenario been validated? Who is liable for accidental injury? How do you recover from failures?

Its pros are impressive dynamic capabilities, suitability for motion control and embodied intelligence research, clear direction for open interfaces leaving room for secondary development, and joint/cooling designs geared toward long-term debugging rather than just posing. The cons are direct: high barriers for beginners and non-technical teams, an engineering gap between demo capabilities and stable work, and dirty work like safety, task management, and data logging that you must fill yourself.

Recommended for university labs, embodied intelligence algorithm teams, and companies with industrial scenarios willing to do system integration. Not recommended for those who just want an office display piece or non-technical teams expecting immediate replacement of human labor.

Actionable advice: Don't just watch backflip videos. Ask about a minimal task: Walk from Point A to Point B, stop when encountering obstacles, recover after manual emergency stop, and export logs. If you can smooth out this process, then talk about deployment. The selling point of the PM01 is its potential as a development platform. Whether the platform becomes a product depends on your team filling in the gaps for safety, workflows, and validation.