Legged Robot State Estimation via Dual Kalman Filter Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing state estimation methods for legged robots suffer from error accumulation and inaccuracy due to foot end slipping, deformation, and sensor noise, especially during long-term motions, which affects the precision of position and pose information.
Innovation Solution
The method employs two Kalman filters to fuse sensor information from sensors operating at different frequencies, using historical state information to correct and update current state estimates, thereby reducing error accumulation and improving real-time state estimation precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If body sensing-based sensors (IMU, leg encoder) are used for state estimation, then real-time state estimation can be obtained, but error accumulation occurs during long-term motions
Solution Approach 1:
The patent implements feedback by using the visual sensor to periodically correct the state estimation results from body sensing-based sensors. The visual sensor provides absolute position and pose information that serves as feedback to eliminate cumulative errors, while the body sensing sensors provide continuous real-time data between visual corrections.
Solution Approach 2:
The patent introduces a visual sensor as an intermediary to bridge the gap between body sensing-based estimation and ground truth. The visual sensor acts as a mediator that provides periodic correction data to compensate for drift and error accumulation in the inertial and encoder-based estimation system.
2Adaptability or versatility
If multi-sensor fusion estimation is used, then state estimation can be performed through various means, but sensor noise and drift affect estimation results
Solution Approach 1:
The patent segments the sensor system into two distinct groups: body sensing-based sensors (IMU, encoders) for continuous real-time estimation, and visual sensors for periodic correction. This segmentation allows each sensor type to be optimized for its specific function while reducing the impact of individual sensor deficiencies.
Solution Approach 2:
The patent changes the operational parameters of different sensors, using body sensing sensors at high frequencies for real-time tracking and visual sensors at lower frequencies for correction. This parameter differentiation allows the system to leverage the strengths of each sensor type while minimizing their weaknesses.
3Productivity
If body sensing-based sensors are used, then real-time control data can be obtained, but foot end slipping and deformation cause estimation errors
Solution Approach 1:
The visual sensor serves as an intermediary that provides reliable external reference information to correct errors introduced by foot end slipping and deformation. The visual system observes the robot's position and pose from an external frame of reference, independent of mechanical contact issues.
Solution Approach 2:
The patent implements feedback by comparing the state estimation from body sensing sensors with visual measurements and using the visual data to correct systematic errors caused by foot slipping and deformation. This feedback loop continuously refines the reliability of state estimation.
Data Source
AI summary
In a state estimation method for a legged robot, first sensor information and second sensor information of the legged robot are received. First state information of the legged robot for a period of time is determined, via a first Kalman filter, based on the first sensor information and the second sensor information. Third sensor information of the legged robot is received. Second state information of the legged robot is determined, via a second Kalman filter, based on the third sensor information and the first state information for the period of time. First state information of the legged robot at a current time is updated based on the second state information of the legged robot, to determine state information of the legged robot at the current time.