Abstract-- Differentiable simulators provide analytic gradients, enabling more sample-efficient learning algorithms and paving the way for data intensive learning tasks such as learning from images. In this work, we demonstrate that locomotion policies trained with analytic gradients from a differentiable simulator can be successfully transferred to the real world. Typically, simulators that offer informative gradients lack the physical accuracy needed for sim-to-real transfer, and viceversa. A key factor in our success is a smooth contact model that combines informative gradients with physical accuracy, ensuring effective transfer of learned behaviors. To the best of our knowledge, this is the first time a real quadrupedal robot is able to locomote after training exclusively in a differentiable simulation. The majority of Reinforcement Learning (RL) algorithms rely on Zeroth-order Gradient (ZoG) estimates during optimization, allowing the use of conventional physics simulators that are typically non-differentiable.

Reinforcement learning (RL) often necessitates a meticulous Markov Decision Process (MDP) design tailored to each task. This work aims to address this challenge by proposing a systematic approach to behavior synthesis and control for multi-contact loco-manipulation tasks, such as navigating spring-loaded doors and manipulating heavy dishwashers. We define a task-independent MDP to train RL policies using only a single demonstration per task generated from a model-based trajectory optimizer. Our approach incorporates an adaptive phase dynamics formulation to robustly track the demonstrations while accommodating dynamic uncertainties and external disturbances. We compare our method against prior motion imitation RL works and show that the learned policies achieve higher success rates across all considered tasks. These policies learn recovery maneuvers that are not present in the demonstration, such as re-grasping objects during execution or dealing with slippages. Finally, we successfully transfer the policies to a real robot, demonstrating the practical viability of our approach.

The emergence of differentiable simulators enabling analytic gradient computation has motivated a new wave of learning algorithms that hold the potential to significantly increase sample efficiency over traditional Reinforcement Learning (RL) methods. While recent research has demonstrated performance gains in scenarios with comparatively smooth dynamics and, thus, smooth optimization landscapes, research on leveraging differentiable simulators for contact-rich scenarios, such as legged locomotion, is scarce. This may be attributed to the discontinuous nature of contact, which introduces several challenges to optimizing with analytic gradients. The purpose of this paper is to determine if analytic gradients can be beneficial even in the face of contact. Our investigation focuses on the effects of different soft and hard contact models on the learning process, examining optimization challenges through the lens of contact simulation. We demonstrate the viability of employing analytic gradients to learn physically plausible locomotion skills with a quadrupedal robot using Short-Horizon Actor-Critic (SHAC), a learning algorithm leveraging analytic gradients, and draw a comparison to a state-of-the-art RL algorithm, Proximal Policy Optimization (PPO), to understand the benefits of analytic gradients.

Loco-manipulation planning skills are pivotal for expanding the utility of robots in everyday environments. These skills can be assessed based on a system's ability to coordinate complex holistic movements and multiple contact interactions when solving different tasks. However, existing approaches have been merely able to shape such behaviors with hand-crafted state machines, densely engineered rewards, or pre-recorded expert demonstrations. Here, we propose a minimally-guided framework that automatically discovers whole-body trajectories jointly with contact schedules for solving general loco-manipulation tasks in pre-modeled environments. The key insight is that multi-modal problems of this nature can be formulated and treated within the context of integrated Task and Motion Planning (TAMP). An effective bilevel search strategy is achieved by incorporating domain-specific rules and adequately combining the strengths of different planning techniques: trajectory optimization and informed graph search coupled with sampling-based planning. We showcase emergent behaviors for a quadrupedal mobile manipulator exploiting both prehensile and non-prehensile interactions to perform real-world tasks such as opening/closing heavy dishwashers and traversing spring-loaded doors. These behaviors are also deployed on the real system using a two-layer whole-body tracking controller.