State-Dependent Sensor Measurement Model for Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing state estimation methods for mobile robots fail to accurately account for varying sensor measurement noise and bias, which can lead to hazardous situations due to noisy and uncertain sensor readings, as they do not stochastically estimate sensor bias and noise based on the robot's and environment's states.
Innovation Solution
Implementing a state-dependent sensor measurement model using a mixture density network (MDN) that learns to estimate the sensor's measurement probability density function, allowing for the estimation of measurement bias and noise conditioned on the robot's and environment's states, and integrating this model with the Extended Kalman Filter (EKF) for improved localization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional state estimation methods are used, then the system is simple to implement, but measurement precision deteriorates due to inability to account for varying sensor noise and bias
Solution Approach 1:
The patent applies dynamics by transitioning from static measurement noise models to dynamic, state-dependent noise models. The measurement noise covariance matrix Q is made a function of the robot's state x, allowing the system to adaptively account for varying sensor noise and bias conditions in real-time, thereby improving measurement precision while managing complexity through state-aware adaptation
Solution Approach 2:
The patent changes the parameters of the measurement model from fixed values to state-dependent functions. Specifically, the measurement noise covariance Q and bias are transformed into parameters that vary with the robot's state x, enabling the system to capture the stochastic nature of measurement errors and improve estimation accuracy under different operational conditions
2Measurement precision
If state-dependent measurement models are implemented, then measurement precision improves, but computational complexity increases
Solution Approach 1:
The patent implements feedback by continuously using the estimated robot state to adjust the measurement noise model parameters. The state-dependent covariance matrix Q(x) and bias models receive feedback from the current state estimation, allowing the system to adaptively compensate for noise and bias variations without requiring complex offline computations, thus balancing precision improvement with computational feasibility
Data Source
AI summary
A noise and bias can be determined for a sensor. An input vector can be received. A parameter vector can be generated based at least in part on a feed-forward neural network. Components can be determined using the parameter vector based at least in part on a mixture model. A conditional probability density function can be generated based at least in part on the conditional probability density function.