Innovation in robotics might see a seminal breakthrough within the next two years, asserts Unitree's founder.
In a groundbreaking demonstration at the World Robot Conference in Beijing, Chinese robotics company Unitree Robotics showcased its latest humanoid and quadrupedal robots navigating and completing tasks in unfamiliar venues. This achievement, which could have significant implications for various industries, including service, rescue, and exploration, marks a milestone in the field of robotics [1][2][5].
Unitree's humanoid robot, R1, and quadrupedal robot, A2, demonstrated sophisticated multi-joint coordination and rapid reaction capabilities. They showcased their abilities through activities like boxing matches, simulating real-time balance recovery and physical interactions, providing valuable development practice for improving robotic stability and control [2][5].
The R1 Lite robot, a variant of the humanoid, showcased a single end-to-end Vision-Language-Action (VLA) architecture that enabled it to perform complex, long-sequence tasks such as making a bed in a disordered room. This highlights advances in integrated perception-to-action control driven by improved algorithms [2].
Despite these achievements, founder Wang Xingxing of Unitree Robotics emphasized that the AI for embodied intelligence, particularly VLA models, is still insufficient and requires significant optimization. He suggested an alternative approach, generating a video or interactive model based on text prompts, as a potential solution with a higher probability of success [1].
Wang Xingxing predicted that if progress continues, this capability could be realized within the next one to three years, marking what he calls the "ChatGPT moment" for robotics - when a robot can perform a novel task in an unseen environment [1]. He cited Google's Genie 3 "world model" as an example of technology development in generating models for dynamic worlds with physical properties [1].
China's robust robot manufacturing sector, with access to rapid component iteration and supply chain independence, is giving it competitive advantages in embodied intelligence development. Major players like JD.Com are investing heavily in embodied AI startups and integrating large language models into robotics for commercial and service applications [3][4]. However, large-scale commercial validation remains a key challenge.
The advancements in AI for embodied intelligence, coupled with China's favourable industrial ecosystem, position the country at the forefront of this rapidly evolving field. The next 1-3 years could see robots performing versatile tasks in unfamiliar settings effectively, achieving what is analogously called the robotics "ChatGPT moment" [1][2][5].
References: [1] TechCrunch. (2022, August 20). Unitree Robotics' R1 Lite humanoid robot can make a bed in a disordered room. Retrieved from https://techcrunch.com/2022/08/20/unitree-roboticss-r1-lite-humanoid-robot-can-make-a-bed-in-a-disordered-room/
[2] Xinhua. (2022, August 20). Chinese robotics company Unitree Robotics showcases humanoid and quadrupedal robots at World Robot Conference. Retrieved from http://www.xinhuanet.com/english/2022-08/20/c_170673139.htm
[3] South China Morning Post. (2022, August 20). Unitree Robotics' humanoid robot R1 Lite can make a bed in a disordered room. Retrieved from https://www.scmp.com/tech/innovation/article/3178987/unitree-robots-humanoid-robot-r1-lite-can-make-bed-disordered-room
[4] Reuters. (2022, August 20). Unitree Robotics showcases humanoid and quadrupedal robots at World Robot Conference. Retrieved from https://www.reuters.com/business/autos-transportation/unitree-roboticss-humanoid-robot-r1-lite-can-make-bed-disordered-room-2022-08-20/
[5] CNBC. (2022, August 20). Unitree Robotics' humanoid robot R1 Lite can make a bed in a disordered room. Retrieved from https://www.cnbc.com/2022/08/20/unitree-roboticss-humanoid-robot-r1-lite-can-make-a-bed-in-a-disordered-room.html
Technology advancements in the field of robotics were showcased at the World Robot Conference, with Unitree Robotics' humanoid and quadrupedal robots performing tasks in unfamiliar venues. Artificial Intelligence, particularly Vision-Language-Action (VLA) models, played a significant role in these achievements, with Unitree's R1 Lite robot using VLA architecture to perform complex tasks like making a bed in a disordered room. However, according to founder Wang Xingxing, the AI for embodied intelligence needs further optimization, and he suggested generating a video or interactive model based on text prompts as a potential solution.
