Robot Inertial Parameter Estimation for Closing the Simulation Reality Gap
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is a significant 'reality gap' between simulated and real-world robot environments due to inaccuracies in physical parameters, simplified contact models, ignored hardware limitations, noise, and latency, causing control and learning algorithms to perform poorly in real-world scenarios.
Innovation Solution
The technique modifies inertial parameters in virtual robot models by using physical measurements from real-world robots to extract optimized inertial properties through finite differencing, median filtering, and nonlinear least squares optimization, decoupling them from control limitations and friction, and applying recursive inverse dynamics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If simplified contact models and inaccurate physical parameters are used in simulation, then computational complexity is reduced, but the accuracy of robot dynamic behavior modeling deteriorates
Solution Approach 1:
The patent applies parameter changes by optimizing inertial parameters (mass, center of mass, inertia tensor) through nonlinear least squares optimization. The system adjusts these parameters to minimize the difference between simulated and measured torques, thereby improving accuracy without increasing model complexity. This directly resolves the contradiction by finding optimal parameter values that maintain simplicity while achieving high precision.
Solution Approach 2:
The patent replaces complex mechanical measurement systems with a computational approach using recursive inverse dynamics and optimization algorithms. Instead of using complex physical measurement devices to directly obtain inertial properties, the system substitutes this with a software-based method that computes inertial parameters from torque and motion data, reducing overall system complexity while improving accuracy.
2Ease of operation
If finite differencing is used to extract joint velocities and accelerations, then motion data can be obtained from position measurements, but noise is introduced to the system
Solution Approach 1:
The patent introduces an intermediary filtering step between finite differencing and the optimization process. A median filter is applied to the computed velocities and accelerations to remove noise introduced by finite differencing. This intermediary processing step maintains the ease of motion data extraction while significantly improving signal reliability by eliminating noise before the data is used in torque computation and optimization.
3Measurement precision
If inertial properties are decoupled from control limitations and friction, then estimation accuracy is improved, but the problem formulation becomes more complex
Solution Approach 1:
The patent extracts and isolates inertial properties from other dynamic effects such as friction and control limitations. By formulating the optimization problem to specifically solve for inertial parameters independently, the system achieves higher estimation accuracy. The nonlinear least squares framework allows inertial properties to be determined separately from other factors, maintaining formulation complexity only where necessary for accurate extraction.
4Productivity
If recursive inverse dynamics is used for torque computation, then computational intensity is reduced, but the requirement for accurate motion data increases
Solution Approach 1:
The patent applies preliminary filtering and processing to motion data before it is used in recursive inverse dynamics. By pre-processing the position data through finite differencing and median filtering to obtain clean velocity and acceleration estimates, the system ensures high-quality input data for the torque computation. This preliminary action maintains computational efficiency while guaranteeing the accuracy requirements are met.
Data Source
AI summary
Methods and systems for modifying the inertial parameters used in a virtual robot model that simulates the interactions of a real-world robot with an environment to better reflect the actual inertial properties of the real-world robot. In one aspect, a method includes obtaining joint physical parameter measurements for the joints of a real-world robot, determining simulated joint physical parameter values for each of the joint physical parameter measurements, and adjusting an estimate of inertial properties of the real-world robot used by the virtual robot dynamic model to reduce a difference between the simulated joint physical parameter values and the corresponding joint physical parameter measurements.