Pose Estimation of a Thruster-Driven Bioinspired Multi-Link Robot

2025-10-01

AI Summary

This work addresses pose and configuration estimation for underactuated, biomimetic multilink propulsive robots with extreme sensing constraints—only one gyroscope per link. Method: We propose an unscented Kalman filter integrated with Gaussian process residual learning to jointly estimate global pose and configuration. The method explicitly models non-zero-mean, non-Gaussian noise to enhance robustness; critically, we identify overlapping input spaces across distinct gaits and, for the first time, enable cross-gait shared training data—significantly reducing data requirements. Contribution/Results: Hardware experiments demonstrate reliable convergence under free-floating conditions. A single cross-gait model achieves pose estimation accuracy comparable to gait-specific models trained on large datasets, while reducing training data volume by ~60%. Our core contribution is high-precision pose estimation under minimal sensing, enabled by a novel gait-generalization mechanism that overcomes traditional data bottlenecks in underactuated robotic perception.

Technology Category

• Intelligent Robots: State Estimation
• Machine Learning: Multimodal Learning
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

This work demonstrates pose (position and shape) estimation for a free-floating, bioinspired multi-link robot with unactuated joints, link-mounted thrusters for control, and a single gyroscope per link, resulting in an underactuated, minimally sensed platform. Through a proof-of-concept hardware experiment and offline Kalman filter analysis, we show that the robot's pose can be reliably estimated. State estimation is performed using an unscented Kalman filter augmented with Gaussian process residual learning to compensate for non-zero-mean, non-Gaussian noise. We further show that a filter trained on a multi-gait dataset (forward, backward, left, right, and turning) performs comparably to one trained on a larger forward-gait-only dataset when both are evaluated on the same forward-gait test trajectory. These results reveal overlap in the gait input space, which can be exploited to reduce training data requirements while enhancing the filter's generalizability across multiple gaits.

Problem

• Estimating pose for underactuated bioinspired robots with minimal sensing
• Compensating non-Gaussian noise using Gaussian process residual learning
• Reducing training data needs while maintaining multi-gait performance

Innovation

• Unscented Kalman filter with Gaussian process residual learning
• Compensating for non-Gaussian noise in thruster-driven robot
• Multi-gait training enhances filter generalization across gaits