CTBC: Contact-Triggered Blind Climbing for Wheeled Bipedal Robots with Instruction Learning and Reinforcement Learning

TL;DR

This work tackles the challenge of robust obstacle traversal for wheeled-bipedal robots with varying wheel sizes and tire types by introducing CTBC, a contact-triggered blind climbing framework that blends feedforward instruction learning with reinforcement learning. The method uses an asymmetric actor–critic policy and a contact-based trigger to switch between rolling and leg lifting, enabling stairs and gaps to be crossed using proprioception alone. Extensive sim-to-real validation on LimX Tron1 and Cowarobot R0, including ablations and real-world tests up to 20 cm steps, demonstrates strong cross-platform generalization and zero-shot transfer. The results indicate that combining guided lifting primitives with contact-triggered control yields robust, scalable climbing capabilities, with future work aimed at reducing gait bias and integrating perception for full autonomous navigation.

Abstract

In recent years, wheeled bipedal robots have garnered significant attention due to their exceptional mobility on flat terrain. However, while stair climbing has been achieved in prior studies, these existing methods often suffer from a severe lack of versatility, making them difficult to adapt to varying hardware specifications or diverse complex terrains. To overcome these limitations, we propose a generalized Contact-Triggered Blind Climbing (CTBC) framework. Upon detecting wheel-obstacle contact, the framework triggers a leg-lifting motion integrated with a strongly-guided feedforward trajectory. This allows the robot to rapidly acquire agile climbing skills, significantly enhancing its capability to traverse unstructured environments. Distinct from previous approaches, CTBC demonstrates superior robustness and adaptability, having been validated across multiple wheeled bipedal platforms with different wheel radii and tire materials. Real-world experiments demonstrate that, relying solely on proprioceptive feedback, the proposed framework enables robots to achieve reliable and continuous climbing over obstacles well beyond their wheel radius.

Figures (9)

• Figure 1: We have developed a contact-triggered blind climbing control policy that works for wheeled-legged robots of various tire sizes, enabling them to conquer a variety of challenging terrains.
• Figure 2: Overview of our universal contact-triggered blind stair-climbing framework. The overall framework is mainly composed of a state estimator and an asymmetric actor-critic network. For elastic tires, we introduce a contact force sliding window to simulate more realistic contact. When the contact force in the xy direction of the foot exceeds a threshold, it triggers the designed feedforward reference trajectory to guide the robot to lift its leg. After annealing the feedforward trajectory, the method can be zero-shot transferred to the physical robot.
• Figure 3: Terrain type. From left to right and top to bottom: smooth slope, rough slope, discrete obstacles, and stairs.
• Figure 5: When either wheel makes contact, only the contacting leg lifts.
• Figure 6: If both wheels are in contact, the leg with stable contact or the larger contact force is chosen to lift.