Tip-Over Detection and Avoidance of Wheeled Mobile Robots in Unknown Uneven Environments

<p><strong>Wheeled mobile robots (WMRs) have become indispensable in industries such as agriculture, manufacturing, defence, and planetary exploration. While they perform reliably on flat and structured surfaces, their operation in unknown and uneven terrains presents significant challenges. Among these challenges, tip-over hazards pose critical risks, potentially causing mission failures and damage to the robot. This research addresses the limitations of existing methods for tip-over detection and avoidance of robots, by proposing a comprehensive framework that combines practical solutions to enhance the stability of four-wheeled mobile robots (FWMRs) navigating unstructured terrains.</strong></p><p>A review of the existing static measures reveals that Force-Angle Stability Margin (FASM), Tip-Over Moment (TOM) and Moment Height Stability (MHS) prove useful, however, they fail to capture the full range of dynamic effects and terrain interactions relevant to FWMRs in rugged terrains. Moreover, the practical accurate estimation of the stability measures in unknown environments remains unsolved, whereas tip-over avoidance in these environments remains an open problem.</p><p>With the above research gaps in focus, a novel stability measure--Acceleration Angle Tip-Over Measure (AATOM) is proposed, comprehensively considering all the dynamic factors leading to a tip-over specific to FWRMs. Unlike existing methods, this measure quantitatively evaluates tip-over stability, expressed in actionable terms suitable for real-time control systems. The measure is also extended to the unstable region, where a measure of the control actions for recovery (AATOM--R) from an ongoing tip-over is developed.</p><p>To support the practical development of the proposed stability measure, two complementary models are developed. The first is a physics-based model using interoceptive sensors to estimate forces and moments acting on the robot. The second is a deep learning-based predictive model that uses interoceptive sensor data to forecast and estimate the support polygon of the robot. By combining these models, the overall system achieves real-time tip-over stability estimation, even in the absence of prior knowledge about the terrain.</p><p>To effectively avoid tip-overs in harsh unknown environments where there is no map of the terrain, a deep-learning-based stability predictor is developed using the past data acquired by interoceptive sensors. Building on these predictive capabilities, a control strategy is designed to mitigate tip-over risks by adjusting robot trajectories and velocities. In the event of an unpredictable sudden tip-over or an unavoidable tip-over from local trajectory adjustments, the control algorithm uses the recovery measure developed in this thesis to recover from the ongoing tip-over. The integration of this control system into a manually operated FWMR demonstrates its effectiveness in real-world scenarios, significantly reducing tip-over incidents.</p><p>In summary, this work presents a practical implementation of the framework on a skid-steering FWMR equipped with interoceptive sensors. Experimental evaluations validate the system's ability to operate autonomously in unknown environments while maintaining stability. The results highlight the adaptability of the proposed overall approach, where safety and reliability are critical in challenging terrains.</p>