Robot Contact Localization Using Bayes Filtering for Occluded Objects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional localization methods for objects manipulated by robots fail to accurately determine the precise pose and velocity of objects with arbitrary geometries and kinematics, especially in contact situations, due to limitations in camera-based systems and the inability to handle multimodal object distributions in real-time.
Innovation Solution
A probabilistic approach using a Bayes filter algorithm with a particle filter approximation, which estimates the object pose and velocity by combining state transition probabilities from a nondeterministic motion model with measurement probabilities from a measurement model, incorporating physics simulations and impedance control to account for robot and object interactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If camera-based localization methods are used, then object detection is possible in open-loop conditions, but precision fails in contact situations and when objects are obscured
Solution Approach 1:
The patent introduces a probabilistic state estimation system as an intermediary between robot actions and object localization. This system uses Bayes filters and particle filters to process robot configuration measurements and interaction forces, generating accurate object pose estimates even when visual sensors fail due to obscuration or contact conditions
Solution Approach 2:
The patent replaces camera-based optical detection with a mechanics-based localization approach. By using robot configuration measurements, interaction forces, and physics simulations to model object dynamics, the system achieves reliable localization through mechanical interaction data rather than optical observation
2Adaptability or versatility
If conventional localization methods are used, then simple object tracking is possible, but arbitrary robot kinematics and arbitrary object geometries cannot be handled
Solution Approach 1:
The patent creates a universal localization framework that handles arbitrary robot kinematics and object geometries through a single probabilistic state estimation system. The Bayes filter and particle filter algorithms are geometry-agnostic and can process any robot configuration data, making the system universally applicable without requiring geometry-specific customization
Solution Approach 2:
The patent transforms the localization problem from a geometry-dependent visual recognition task into a parameter-based state estimation problem. By representing objects through physical parameters (mass, center of gravity, inertia tensor) and robot configurations, the system adapts to arbitrary geometries through parameter input rather than structural complexity
3Loss of information
If complete object pose including velocity is determined, then full object state is known, but computational complexity increases for real-time operation
Solution Approach 1:
The patent implements a dynamic state estimation system that recursively updates object pose and velocity information as new measurements become available. The particle filter algorithm continuously evolves the probability distribution over object states, maintaining complete information while adapting to changing conditions in real-time through incremental computation rather than batch processing
Solution Approach 2:
The patent uses particle filter approximation to compute only the essential components of the complete object state (pose and velocity) with sufficient accuracy for real-time control. Rather than computing the full probability distribution exactly, the system uses a finite set of particles to represent the state, achieving practical real-time performance while retaining complete state information
Data Source
AI summary
At least one robot is disclosed for object manipulation, on the basis of a probabilistic approach without object detection by visual sensors, a probability distribution is determined approximately by an iterative, recursive application of a Bayes filter algorithm, wherein state transition probabilities obtained, in a nondeterministic motion model for system simulation, are multiplicatively linked with measurement probabilities obtained at the start of the application. For this purpose, a physics simulator which completely includes the physics of the system with respect to forces and dynamics and the physical system relationships resulting therefrom, and a controller, which controls the physics simulator while at the same time the simulation system state is fed back and which influences the compliance of the robot with respect to the robot-object environment on the basis of control variables, are incorporated, and measurement results of axis-specific measurements are taken into account in the measurement model for the plausibility checks.