On April 9th, GAC Group officially launched its third-generation humanoid robot, GoMate, at a press conference on innovative products and services in the Guangdong Province's AI and robotics industry. A team leader from GAC's robotics team revealed that GoMate is already being used in security applications, performing patrol duties, and will expand into automotive production lines and after-sales service markets in the future.
Regarding GAC's foray into humanoid robots, Zhang Aimin explained that new energy vehicles and humanoid robots share significant technological and supply chain overlaps, such as chips, lidar, and vision sensors. Furthermore, automotive production workshops and 4S stores provide ample application scenarios for humanoid robots. GAC Group aims to leverage intelligent robots to deepen its intelligent technology layout, extending its existing technology, supply chain, and brand advantages from the automotive industry to broader fields.
Addressing the challenges facing humanoid robots today—including insufficient AI generalization capabilities, limited movement and manipulation abilities, high costs, and a lack of mature application scenarios—GAC established a dedicated intelligent robotics team. After two years and three months of intensive development and two rounds of optimization iterations, they achieved significant innovations in movement, manipulation, brain technology, and self-developed hardware, culminating in GoMate.
GoMate's most prominent feature is its innovative "wheel-leg combined" configuration design. Zhang Aimin pointed out that this design enhances efficiency, allowing for an energy-saving two-wheel mode on flat surfaces and switching to a more stable four-wheel mode in challenging environments like stairs, grass, and crowded areas.
In terms of its "brain," the GAC team fine-tuned open-source large models for specific scenarios such as public safety. By integrating AI large models and autonomous driving algorithms through a cloud-edge-end collaborative architecture, GoMate is granted decision-making capabilities. Its multi-modal perception technology achieves 360-degree panoramic perception and centimeter-level blind spot detection, facilitating more precise instruction execution. Reinforcement learning algorithms enable GoMate to iterate quickly using low-cost data.
GoMate employs a remote control and visual sharing operation mode, allowing operators to command it to perform complex, high-precision tasks from anywhere. In motion control, GoMate utilizes traditional MPC and WBC whole-body control algorithms, boasting 38 degrees of freedom, enabling stable walking on complex terrain. Its self-developed actuators are only coin-sized, weighing 50 grams, yet capable of providing a continuous current output of 20A.
Another highlight is GoMate's dexterous hand, weighing less than 500 grams but capable of lifting 1.5 kilograms. Its fingertips and knuckles employ a biomimetic structure combining rigidity and flexibility, balancing high load capacity and high compliance, enabling it to grasp objects of various materials. "We have self-developed all our core functional components, and key indicators show strong competitiveness," Zhang Aimin emphasized.
GoMate is not a prototype; GAC Group has a clear mass production plan: this year will see the mass global sale of self-developed components, with GoMate undergoing demonstration applications in various industries. Small-scale production is planned for 2026, gradually scaling up to mass production.