The field of quadrupedal robotics is witnessing significant advancements, with a focus on enabling these robots to perform complex tasks with agility and adaptability. Researchers are exploring innovative approaches to address the challenges posed by quadrupedal robots' nonlinear dynamics and high degrees of freedom. One key direction is the development of learning frameworks that can effectively integrate expert demonstrations and reinforcement learning to achieve sample-efficient imitation learning and diverse skill acquisition. Another area of focus is the improvement of control pipelines for real-time deployment, allowing for smooth gait switching and high-performance locomotion. Noteworthy papers in this area include:

• A paper demonstrating the use of Reverse Curriculum Reinforcement Learning to enable quadrupedal robots to mount skateboards, showcasing robustness to variations in skateboard position and orientation.
• A paper presenting a Multi-Task Learning framework that enables a single neural network to predict actions for multiple locomotion behaviors, achieving high accuracy and simplifying the control pipeline.
• A paper introducing APEX, a simple yet versatile imitation learning framework that integrates demonstrations directly into reinforcement learning, achieving sample-efficient imitation learning and enabling the acquisition of diverse skills.