A state estimation method and system for multi-modal perception of a legged robot

Abstract

The invention discloses a state estimation method and system for multimodal perception of a legged robot. The method includes: measuring the acceleration and angular velocity of the legged robot based on an IMU sensor, and predicting the state mean and covariance of the legged robot Obtain the angular positions of all joints in the footed robot based on the joint encoder, and calculate the measurement value and predicted value corresponding to the foot point position of each foot in the footed robot; based on the measurement value and the obtained The predicted value is used to calculate the measurement residual of the legged robot; based on the measurement residual and the measurement Jacobian, the state mean value and the covariance are corrected to obtain the final state of the legged robot. The method will improve the accuracy and stability of state estimation for legged robots.

Description

technical field [0001] The invention relates to the field of robots, in particular to a state evaluation method and system for multimodal perception of a legged robot. Background technique [0002] The research and development of legged robots is inspired by mammals, which is an important embodiment of bionics and robotics. It has good environmental adaptability, wide range of motion, strong load capacity, and a certain ability to operate independently. Disaster rescue, military reconnaissance and other tasks have received extensive attention. However, the accuracy of the sensor on the ontology is limited, and the measurement data is uncertain, so a good estimation of its state is required to execute closed-loop control instructions and adapt to the non-structural environment. State estimation includes knowledge of the robot's motion and characteristics, fully describing the robot's motion over time in a known environment, and in order to obtain the best estimate with imper...

AI Technical Summary

Problems solved by technology

The traditional state assessment method is to complete the state estimation and measurement model by combining the dynamic model of the robot, using the dynamic model to give the expected state, and using various sensors on the robot to provide one or more state quantities of the robot. The method faces several problems: 1) the dynamics are highly nonlinear, and the calculation of the dynamic model of the numerical solution is not accurate; 2) the state space of the legged robot is very large, including not only the posture of the body, but also the position and position of the foot joints. attitude; 3) footed robots interact with the environment through multiple intermittent ground contacts and impacts, making the sensor noisier

[0003] It can be seen that traditional algorithms usually rely on the data fusion of multiple bodies and external sensors for state estimation. However, external sensors are affected by characteristic factors such as light intensity, distance measurement, and sound amplitude, and their measurement reliability cannot be guaranteed. Reduced accuracy and stability of state estimation

In addition, because legged robots interact with the surrounding environment through intermittent foot-ground contact, additional noise will be introduced in the measurement, and how to minimize the impact of noise is also a challenging problem

Images