Unitree founder Wang Xingxing: Robot core bottleneck is solving millimeter-level errors

On September 24, the 5th Global Digital Trade Expo was held in Hangzhou. Wang Xingxing, founder of Unitree Robotics, delivered a keynote speech titled "From Mechanics to Intelligence — The Evolution of Embodied Intelligence". Staying true to his signature speaking style, he used the company's products as a starting point to systematically review its robotics development.

GD01: The World's First Mass-Produced Manned Robot

At this year's expo, the most eye-catching exhibit at the Unitree booth was the three-meter-tall GD01 manned robot. Officially launched in May this year, it is the world's first mass-produced version of a manned robot.

Addressing widespread curiosity about why Unitree chose to develop a large-scale manned robot, Wang Xingxing responded that as the underlying technologies of AI-driven robotics continue to mature, the development of large-scale robots is not in conflict with other efforts. He described large-scale robots as an unstoppable trend in the industry, adding that even if Unitree had not taken the lead, other companies would have launched similar products within the next few years.

He defined the GD01 as the "off-road vehicle" of the robotics field. Due to its superior mobility, the robot is not intended for urban use but rather for complex outdoor terrains, much like an off-road vehicle.

Progress in AI Models and Real-Time Motion Generation

Wang Xingxing noted that in recent months, Unitree's large models for robotics have made continuous progress. The robots can now generate and execute actions in real time based on human voice commands. However, because both speech understanding and motion generation rely on real-time AI computation, there is still a slight delay during execution.

He stated that Unitree's robots are currently capable of executing long sequences of continuous tasks and possess anti-interference capabilities, though there remains significant room for improvement in operational efficiency. He expressed hope that by the end of this year or next year, any user could issue a command to a robot, and the robot would be able to generate the corresponding action. Once the ability to generate arbitrary actions becomes sufficiently robust, robots will be able to work on a large scale.

The Coming "ChatGPT Moment" for Embodied Intelligence

Reflecting on the development of the AI industry, Wang noted that the emergence of ChatGPT fundamentally reshaped public perception of artificial intelligence and unlocked the industry's imagination. He predicted that the embodied intelligence sector will inevitably experience its own "ChatGPT moment." When robots can complete 80% of tasks in 80% of unfamiliar scenarios using voice-driven embodied capabilities, the industry will reach a critical tipping point for explosive growth. At that stage, global enterprises and resources will flood into the sector, and this inflection point could arrive within the next few years.

Key Challenge: Precision in Real-World Alignment

Wang pointed out that last year, it was already possible to have robots follow voice commands and perform symbolic tasks. However, the biggest current challenge is the insufficient precision in aligning AI model inputs and outputs with the real physical world, resulting in errors of a few millimeters during robot operations.

"Whoever solves this problem in the future will have completely solved the robot problem," Wang said.

Toward Autonomous AI Iteration

Wang also noted that the current development model in the global robotics industry remains relatively primitive. Processes such as code generation, training deployment, real-machine testing, and problem correction rely heavily on human involvement, resulting in low efficiency and insufficient real-world data.

He proposed that traditional manual workflows could be entirely replaced by autonomous AI loops: AI could independently generate code, complete training, deploy to robots, conduct real-machine tests, and autonomously analyze results to drive positive iterative cycles.

This autonomous iteration system would effectively address the core industry pain point of scarce real-world data. For Unitree, while the company could previously test only dozens of robots per day, this AI-driven system would allow it to test hundreds or even thousands of robots daily. The company would only need to scale up production lines to meet demand.

Unitree's Core Value: Driving Industry Progress

Regarding Unitree's development philosophy, Wang Xingxing emphasized that the company's core value does not lie in product sales volume or revenue and profit. Instead, it is about leveraging self-developed products to work with partners and customers in continuously advancing the global high-performance quadruped and humanoid robotics industry.

(Source: Interface News)

Photo credit: Yang Bingke, Interface News