Deep reinforcement learning (RL) uses model-free techniques to optimize task-specific control policies. Despite having emerged as a promising approach for complex problems, RL is still hard to use reliably for real-world applications. Apart from challenges such as precise reward function tuning, inaccurate sensing and actuation, and non-deterministic response, existing RL methods do not guarantee behavior within required safety constraints that are crucial for real robot scenarios. In this regard, we introduce guided constrained policy optimization (GCPO), an RL framework based upon our implementation of constrained proximal policy optimization (CPPO) for tracking base velocity commands while following the defined constraints. We introduce schemes which encourage state recovery into constrained regions in case of constraint violations. We present experimental results of our training method and test it on the real ANYmal quadruped robot. We compare our approach against the unconstrained RL method and show that guided constrained RL offers faster convergence close to the desired optimum resulting in an optimal, yet physically feasible, robotic control behavior without the need for precise reward function tuning.

No abstract

Traditional approaches to quadruped control frequently employ simplified, hand-derived models. This significantly reduces the capability of the robot since its effective kinematic range is curtailed. In addition, kinodynamic constraints are often non-differentiable and difficult to implement in an optimisation approach. In this work, these challenges are addressed by framing quadruped control as optimisation in a structured latent space. A deep generative model captures a statistical representation of feasible joint configurations, whilst complex dynamic and terminal constraints are expressed via high-level, semantic indicators and represented by learned classifiers operating upon the latent space. As a consequence, complex constraints are rendered differentiable and evaluated an order of magnitude faster than analytical approaches. We validate the feasibility of locomotion trajectories optimised using our approach both in simulation and on a real-world ANYmal quadruped. Our results demonstrate that this approach is capable of generating smooth and realisable trajectories. To the best of our knowledge, this is the first time latent space control has been successfully applied to a complex, real robot platform.

We present a control architecture for real-time adaptation and tracking of trajectories generated using a terrain-aware trajectory optimization solver. This approach enables us to circumvent the computationally exhaustive task of online trajectory optimization, and further introduces a control solution robust to systems modeled with approximated dynamics. We train a policy using deep reinforcement learning (RL) to introduce additive deviations to a reference trajectory in order to generate a feedback-based trajectory tracking system for a quadrupedal robot. We train this policy across a multitude of simulated terrains and ensure its generality by introducing training methods that avoid overfitting and convergence towards local optima. Additionally, in order to capture terrain information, we include a latent representation of the height maps in the observation space of the RL environment as a form of exteroceptive feedback. We test the performance of our trained policy by tracking the corrected set points using a model-based whole-body controller and compare it with the tracking behavior obtained without the corrective feedback in several simulation environments. We also show successful transfer of our training approach to the real physical system and further present cogent arguments in support of our framework.

No abstract