Chengdu unveils fifth-generation gongga humanoid robot, claiming leading global brain technology

The Chengdu Humanoid Robot Innovation Center recently introduced the fifth-generation Gongga humanoid robot, asserting that its core brain technology represents the highest level among current global first-tier humanoid robot brains. This advanced system integrates a proprietary embodied intelligence architecture, featuring systemic upgrades across modules such as scene perception, spatio-temporal memory, causal reasoning planning, and action world modeling. The robot establishes a complete technical pipeline spanning structured perception through to execution feedback, enabling end-to-end coordination for environmental understanding, memory updating, task planning, and operation. Key capabilities include building hierarchical 3D scene maps using RGB vision and depth data, allowing it to maintain structured representations of dynamic home environments despite frequent changes. Furthermore, its self-developed physical spatio-temporal memory engine supports long-term identity consistency and multi-round memory alignment, moving beyond the limitations of single-task memory. Decision-making is enhanced by a causal inference framework, which allows the system to autonomously decompose complex tasks and dynamically replan when encountering unexpected environmental changes. The action execution relies on a latent space action conditional world model, enabling the robot to simulate future states and assess risks virtually before executing fine motor tasks like grasping, significantly boosting success rates.