Learning Fault-Tolerant Locomotion with Adaptive Gait Timing

Giovanbattista Gravina, Luca Rossini, Carlo Rizzardo, Arturo Laurenzi, Nikos Tsagarakis

Paper Code

Abstract

Hardware failures require legged robots to rapidly reorganize coordination and gait timing to maintain stability and mobility. This is particularly challenging for larger quadrupeds, where increased mass and tighter actuation limits reduce the feasibility of aggressive, high-frequency compensation strategies often observed on smaller platforms. In this work, we propose a deep reinforcement learning approach for fault-tolerant locomotion under actuator power loss. The method employs an asymmetric actor-critic architecture in which the critic has access to privileged information during training, while the actor learns to reconstruct a corresponding latent representation from proprioceptive observations. We introduce a latent-alignment loss that encourages consistency between actor and critic representations. Additionally, we augment the action space with a learnable gait frequency parameter, enabling adaptive gait timing in response to terrain variations and actuator degradation without predefined faulty-leg strategies. The approach is validated in high-fidelity simulation on uneven terrain and real-world experiments on flat ground using the 68 kg KYON quadruped robot.

Architecture

Project Overview Diagram
The proposed architecture consists of an actor and a critic, each comprising an encoder and a head network. The critic encoder embeds the privileged observation \(\boldsymbol{o}^{\text{p}}\), composed of proprioceptive observation \(\boldsymbol{o}\) and privileged information \(\boldsymbol{e}\), into a latent representation \(\boldsymbol{r}\). This latent vector is then concatenated with \(\boldsymbol{o}\) and the terrain-related observations \(\boldsymbol{o}^{\text{z}}\) to form the input to the critic head for value estimation. In parallel, the actor encoder processes a history of \(H\) proprioceptive observations and outputs a latent vector \(\tilde{\boldsymbol{r}}\). The actor head receives the concatenation of \(\tilde{\boldsymbol{r}}\), \(\boldsymbol{o}\), and \(\boldsymbol{o}^{\text{z}}\) to produce the policy action \(\boldsymbol{a}\).

Simulations

The following MuJoCo simulations show the Kyon quadruped robot, either equipped or not with its upper limbs, tracking velocity commands while powered by our control policies. The quadruped robot is able to climb stairs (left) and handle the upper-body additional payload (right) under joint failure conditions.

We further validate the learned policy sim-to-sim in MuJoCo through the XBot2 middleware, running it on the same software stack deployed on the hardware. The robot traverses slopes (left) and stairs (right).

Experiments

Zero-shot sim-to-real transfer of the learned policy on the hardware.

Full Project Video

Watch the following video for a comprehensive overview of the project.