Self-Position Estimation Using Confidence-Based Sensor Selection

The patent changes the parameter being estimated from raw sensor observation values to state amounts (acceleration bias, gravitational acceleration, speed) that represent the underlying physical conditions. By estimating these parameters and using them to correct sensor readings, the system maintains accuracy even when observation values become constant during uniform motion.

The patent implements feedback by using the estimated state amounts to correct the sensor observation values. The estimated acceleration bias and gravitational acceleration are fed back to adjust the raw acceleration sensor readings, creating a closed-loop system that compensates for sensor errors and maintains estimation accuracy over time.

The patent makes the position estimation system dynamic by adjusting which sensors and state amounts are used based on their confidence levels. When confidence in state amount estimates is low (such as during constant speed linear motion), the system dynamically switches to rely more on image-based estimation and other sensors with higher confidence, rather than continuously using the low-confidence state amounts.

The patent applies local quality by treating different sensors and state amounts differently based on their individual confidence levels. Instead of uniformly using all available data, the system selectively weights or excludes specific sensors and state amounts depending on their current reliability, allowing high-confidence data to dominate the estimation while low-confidence data is minimized or excluded.