Abstract
Mecanum Wheeled Mobile Robots (MWMRs) are increasingly utilized in complex logistics and medical environments due to their omnidirectional mobility. However, their performance is highly dependent on the precise synchronization of all four actuators. This paper proposes a robust active Fault Tolerant Control (FTC) strategy designed to handle both complete motor failures and partial torque degradation. By integrating a Bayesian posterior probability framework into a dual-loop control architecture, the system learns fault parameters in real-time and dynamically blends control laws. Simulation results indicate that this approach maintains precise trajectory tracking and operational safety even under multiple simultaneous actuator compromises, outperforming traditional PID and adaptive control schemes.
Introduction
The unique wheel design of Mecanum robots allows for holonomic movement, enabling them to translate in any direction without changing orientation. While this agility is a significant advantage in constrained spaces, it introduces a critical vulnerability: the robot's stability relies on a delicate balance of forces from all four wheels. If a single motor stalls or experiences a partial loss of torque—common occurrences due to mechanical wear or varying surface friction—this balance is disrupted, leading to significant trajectory drift or total mission failure.
Existing research typically addresses this through either passive or active FTC. Passive strategies are inherently robust but often result in conservative, sluggish performance. Active strategies are more responsive but frequently limited to "binary" fault models, assuming a motor is either fully functional or completely failed. This paper addresses the critical gap of partial actuator failures, providing a more nuanced and practical control solution for real-world operations.
Methodology
The researchers developed a hierarchical two-loop control architecture to manage both high-level navigation and low-level actuator compensation.
Kinematics Loop
The outer loop focuses on path planning and state estimation. An Extended Kalman Filter (EKF) processes sensor data to estimate the robot's current position and orientation. This information is fed into a Model Predictive Control (MPC) solver, which calculates the optimal velocities required to follow the desired trajectory.
Dynamics Loop and Bayesian Learning
The inner loop handles the distribution of torque to the individual wheels. Unlike standard controllers that assume a healthy system, this loop maintains a finite set of "fault hypotheses" representing various states of actuator health (e.g., Wheel 1 at 50% capacity, Wheel 2 at 0%, etc.).
As shown in the architecture diagram above, the system utilizes Bayesian Parameter Learning to calculate the posterior probability of each hypothesis based on real-time deviations from expected movement. The final control law is a weighted sum of individual laws corresponding to these hypotheses. This "probability-weighted" approach prevents the jerky movements associated with hard-switching between controllers and allows the robot to utilize the remaining power of partially degraded actuators.
Performance Evaluation and Results
The proposed FTC strategy was evaluated through high-fidelity numerical simulations across several demanding scenarios, including sequential single-fault events, simultaneous two-fault failures, and obstacle avoidance tasks.
Trajectory Tracking Accuracy
In a complex figure-eight tracking test, the FTC system demonstrated superior precision compared to standard PID and Adaptive MPC (APT) methods. The Bayesian framework allowed for significantly faster identification of fault parameters, enabling the robot to correct its path almost immediately after a failure occurred.
As illustrated in the tracking results, the FTC trajectory (red) remains tightly aligned with the desired path (blue). In contrast, the PID controller (purple) exhibits severe oscillations and drift, while the APT method (yellow) shows a slower recovery time. The red circles at 10s and 20s mark the points where faults were introduced, highlighting the system's rapid adaptation.
Safety and Redundancy
A key insight from the study is the management of redundancy limits. While a Mecanum robot can maintain trajectory with one failed wheel, it becomes underactuated with two. The proposed FTC is specifically optimized to maintain safety and collision avoidance even in these high-stress states, successfully navigating obstacle courses where baseline methods drifted into "danger zones."
Conclusion and Future Work
This research demonstrates that explicitly modeling partial failures and using a Bayesian "soft-switch" significantly improves the reliability of Mecanum robots. By avoiding overly conservative passive designs and binary active models, the proposed FTC maximizes the utility of the robot's remaining hardware.
Future work will focus on transitioning these simulation results to physical hardware. This will involve addressing real-world complexities such as non-linear tire friction and battery voltage fluctuations. Additionally, the researchers aim to expand the finite fault set to a continuous estimation model to further reduce residual errors in cases where failures do not align perfectly with predefined hypotheses.
Summary updated