IT Home On November 11, Chengdu Humanoid Robot Innovation Center released the fifth-generation Gongga humanoid robot, claiming to have the “world’s most advanced core technology for humanoid robot brains”.
As a masterpiece of the embodied intelligence brain technology at the Chengdu Humanoid Robot Innovation Center, the fifth-generation Gongga is equipped with a self-developed integrated embodied intelligence system. It has undergone systematic iterations in core modules such as scene perception, time-space memory, causal reasoning and planning, and action world model, claiming to “represent the highest level of humanoid robot brain technology in the global first tier at present”.
The embodied brain of the Fifth Generation Gongga has established a complete technical chain of “structured perception — temporal and spatial memory — causal planning — action inference — execution feedback”, achieving end-to-end collaboration in environmental cognition, memory updating, task planning, and operational execution.
In the environmental perception dimension, the robot integrates RGB vision, depth information, and pose data to construct a hierarchical 3D scene map. By relying on cross-view entity associations, semantic relationship modeling, and incremental scene update mechanisms, it continuously structures the dynamic home environment. This enables stable recognition of object locations, categories, and placement relationships, adapting to frequent changes such as object relocation and the addition of items in the home environment.
In terms of the memory system, the innovation center’s self-developed physical spacetime memory engine stores and manages spatial coordinates, physical states of objects, time evolution patterns, and historical interaction experiences. The robot can maintain the identity consistency of objects over a long period, supporting temporal memory retrieval, multi-round memory alignment, and automatic conflict resolution. In simple terms, it can “remember” where things were placed, when they were moved, and the outcomes of past operations, overcoming the shortcoming of traditional robots having no memory for single tasks, and enabling long-term household tasks that last for hours or days.
The decision-making and planning module introduces a causal association reasoning framework, no longer relying solely on statistical data to make action choices. The system automatically analyzes the causal dependencies within tasks, preconditions for actions, expected outcomes, and the robot’s own resource constraints, and autonomously breaks down hierarchical tasks; when there are sudden environmental changes or tasks encounter obstacles, it can dynamically replan in real time. It can independently judge and adjust strategies in case of unexpected situations, rather than simply stopping the task.
The action execution level is the core highlight of the Fifth Generation Gongga: it features a submersible space action condition world model. The robot can predict multiple future states, simulate candidate actions, and assess risk assessments in advance within a virtual space; then, by combining motion trajectory optimization and closed-loop predictive control, it can real-time correct operational deviations, significantly improving the success rate of precise household operations such as grasping, positioning, and wiping, while reducing errors like collisions and items falling.
IT Home note: The full name of Chengdu Humanoid Robot Innovation Center is Chengdu Humanoid Robot Innovation Center Co., Ltd. It was established on April 11, 2024. It is the first project under Sichuan Province’s “Number One Innovation Project” and the first humanoid robot innovation center in the central and western regions.