Adaptive energy-regularization paper trains efficient gait transitions on ANYmal-C

Key points

• The paper proposes an adaptive energy reward for reinforcement-learning locomotion so quadrupeds can choose efficient gaits across different speeds.
• The authors demonstrate autonomous gait transitions on ANYmal-C and Unitree Go1, including stable four-beat walking at lower speed and trotting at higher speed.
• It adds meaningful ANYmal research context around energy efficiency and gait selection, not only terrain robustness.