Slip-aware modeling and experimental evaluation of wheeled–legged robots using image processing and learning-based control

Background: Slip and dynamic uncertainties significantly degrade the motion control performance of wheeled–legged robots, especially under varying terrain and load redistribution caused by manipulator motion; Objective: to develop an integrated framework for slip-aware modeling, real-time slip estimation, and learning-based control for wheeled–legged robots. 

Methods: A comprehensive kinematic and dynamic model including explicit longitudinal and lateral wheel slip was formulated, a vision-based localization and slip estimation scheme was implemented using an external camera and concentric circular markers, IMU measurements were used for motion characterization and comparison, and a PID + DDPG control strategy was integrated and evaluated through experiments and simulations. 

Results: The proposed framework reduced trajectory tracking error, improved motion stability under slip, produced smoother torque profiles, and showed better pose estimation accuracy with image processing than inertial sensing alone. 

Conclusion: Combining slip-aware modeling, vision-based estimation, and learning-based adaptive control provides a robust and practical solution for wheeled–legged robot motion in slip-prone environments.