Quadruped robot should be able to “observe” and “think”

The reporter learned from the second Department of the second Academy of China Aerospace Science and Industry Group on the 26th that the research work of the quadruped robot intelligent sensing system of the group intelligent sensing innovation center of the Department has made a breakthrough, which can realize the autonomous positioning and map building function based on the vision scheme, and can map the area of 1000 square meters in one minute, with a high level of intelligence and practicality.

It is reported that this system, like the eyes and brain of a robot, will be carried on a domestic quadruped robot. At present, the robot has the functions of human following, automatic obstacle avoidance and environment modeling.

Guo Rui, head of the system development team, said that climbing stairs and crossing obstacles are difficult for traditional wheeled robots, while quadruped robots can easily cope with complex terrain and have great application potential in exploration and rescue, anti-terrorism and explosion-proof scenes. To build the intelligent perception system, it is necessary to give the quadruped robot the ability of “observation” and “planning and thinking”.

However, the flexible motion mode of quadruped robot, the undulating vibration of the body and the high dynamic environment will greatly reduce the imaging effect of the visual sensing system. In this regard, relying on key technologies such as autonomous and controllable deep learning algorithm, smart chip, slam (synchronous positioning and map building), the development team proposed a low-cost multi-mode fusion intelligent sensing system scheme combining binocular camera, inertial measurement unit and other sensors, which can realize autonomous positioning and three-dimensional semantic map construction in the high dynamic environment of violent fluctuation and vibration of the fuselage Building function, real-time composition of 20 frames per second, map resolution accuracy of centimeter level. This breakthrough significantly improves the intelligent and practical level of quadruped robot.

(editor in chief: Zhao Zhuqing, LV Qian)