Particle Filter Contact State Estimation for Robot Hands
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous dexterous manipulation in robotics is challenging due to difficulties in monitoring the contact state between a robot hand and an object, especially with occlusions occurring during manipulation, and existing particle filtering approaches have limited applications in this domain.
Innovation Solution
A method using particle filters that incorporates a motion model and a measurement model to estimate the contact state between a robot hand and an object, utilizing force or haptic sensors to identify contact positions, velocities, and forces, with measurement models approximated as Gaussian distributions or radial basis functions, and analytically calculated when necessary.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If visual monitoring using a camera mounted on a pan-tilt platform is used to monitor hand-object contact, then the system can observe the contact state, but occlusions occur during manipulation making localization of the object feature being contacted more difficult
Solution Approach 1:
The patent replaces the visual optical system (camera on pan-tilt platform) with a tactile sensing system using force sensors and particle filters. This substitution eliminates occlusion problems by directly measuring contact forces at the hand-object interface, providing accurate contact state information without line-of-sight requirements.
Solution Approach 2:
The patent introduces force sensors as intermediary devices between the robot hand and object to indirectly measure contact state. These sensors act as mediators that translate physical contact into measurable signals, enabling contact state estimation without direct visual observation of the contact point.
2Measurement precision
If tactile sensors located near the point of contact are used to localize hand-object state, then localization accuracy is improved, but the device complexity increases due to sensor integration requirements
Solution Approach 1:
The patent makes the force sensors serve multiple functions: they simultaneously measure contact position, contact force magnitude, and contact normal direction. This multi-functionality reduces the need for separate sensor systems while maintaining high localization accuracy, thereby managing device complexity.
Solution Approach 2:
The patent creates a virtual model (particle filter representation) of the contact state that mirrors the physical contact situation. This computational copy allows the system to estimate contact parameters without requiring complex physical sensor arrays, reducing hardware complexity while maintaining measurement precision.
3Measurement precision
If particle filtering is applied to dexterous manipulation tasks, then contact state estimation accuracy is improved, but the computational complexity increases compared to traditional methods
Solution Approach 1:
The patent segments the contact state estimation problem into distinct components (contact position, contact force, contact normal) that can be estimated independently through the particle filter. This segmentation allows the complex estimation problem to be broken down into manageable parts, reducing overall computational complexity.
Solution Approach 2:
The patent applies particle filtering selectively to estimate only the critical contact parameters needed for manipulation, rather than attempting to model the entire hand-object system state. This partial application reduces computational burden while maintaining sufficient accuracy for the manipulation task.
Data Source
AI summary
A method for identifying the location, orientation and shape of an object that a robot hand touches that includes using a particle filter. The method includes defining an appropriate motion model and a measurement model. The motion model characterizes the motion of the robot hand as it moves relative to the object. The measurement model estimates the likelihood of an observation of contact position, velocity and tactile sensor information given hand-object states. The measurement model is approximated analytically based on a geometric model or based on a corpus of training data. In either case, the measurement model distribution is encoded as a Gaussian or using radial basis functions.