In a remarkable convergence of robotics and biology, researchers at EPFL (École Polytechnique Fédérale de Lausanne) have achieved a groundbreaking milestone – a four-legged robot that can spontaneously switch between walking, trotting, and pronking to navigate challenging terrains. This feat, accomplished through the power of machine learning, offers unprecedented insights into the intricate world of animal locomotion while paving the way for agile robots capable of traversing treacherous environments.

The study, led by the BioRobotics Laboratory in EPFL's School of Engineering, employed deep reinforcement learning (DRL), a form of machine learning, to train the quadruped robot. Through this process, the robot remarkably learned to transition from trotting to pronking – a distinctive leaping gait characterized by an arch-backed posture, commonly observed in animals like springbok and gazelles.

This transition enabled the robot to navigate terrain with gaps ranging from 14 to 30 cm, a formidable challenge that highlights the robot's exceptional agility and adaptability. The research not only pushes the boundaries of robotics but also sheds light on the intricate mechanisms underlying gait transitions in animals.

"Previous research has introduced energy efficiency and musculoskeletal injury avoidance as the two main explanations for gait transitions. More recently, biologists have argued that stability on flat terrain could be more important," explains Ph.D. student Milad Shafiee, the first author of the paper published in Nature Communications. "But animal and robotic experiments have shown that these hypotheses are not always valid, especially on uneven ground."

Shafiee, along with co-authors Guillaume Bellegarda and BioRobotics Lab head Auke Ijspeert, sought to test a novel hypothesis: viability, or fall avoidance, as the driving force behind gait transitions. By employing DRL to train the quadruped robot to cross various terrains, they made a remarkable discovery.

On flat terrain, the researchers observed that different gaits exhibited varying levels of robustness against random pushes, prompting the robot to switch from a walk to a trot – mimicking the behavior of quadruped animals when they accelerate. However, when confronted with successive gaps in the experimental surface, the robot spontaneously transitioned from trotting to pronking, a gait that enabled it to avoid falls more effectively.

"We showed that on flat terrain and challenging discrete terrain, viability leads to the emergence of gait transitions, but that energy efficiency is not necessarily improved," Shafiee explains. "It seems that energy efficiency, which was previously thought to be a driver of such transitions, may be more of a consequence. When an animal is navigating challenging terrain, it's likely that its first priority is not falling, followed by energy efficiency."

To model locomotion control, the researchers drew inspiration from the interplay between the brain, spinal cord, and sensory feedback that drives animal movement. They used DRL to train a neural network to mimic the spinal cord's transmission of brain signals to the body as the robot crossed the experimental terrain.

Through a series of computer simulations, the team assigned different weights to three possible learning goals: energy efficiency, force reduction, and viability. Remarkably, viability emerged as the sole factor that prompted the robot to automatically change its gait, without explicit instruction from the researchers.

The team emphasizes that their observations represent not only the first learning-based locomotion framework in which gait transitions emerge spontaneously during the learning process but also the most dynamic crossing of large consecutive gaps for a quadrupedal robot.

"Our bio-inspired learning architecture demonstrated state-of-the-art quadruped robot agility on the challenging terrain," Shafiee says.

Looking ahead, the researchers aim to expand their work by conducting additional experiments with different types of robots in a wider variety of challenging environments. Their ultimate goal is twofold: to further elucidate the intricate mechanisms underlying animal locomotion and to enable the widespread use of robots for biological research, reducing reliance on animal models and the associated ethical concerns.

This groundbreaking research stands as a testament to the power of interdisciplinary collaboration, where robotics and biology converge to unlock new frontiers in our understanding of movement and agility. As the boundaries between artificial and natural systems blur, the insights gained from this study hold the potential to revolutionize both fields, paving the way for more agile and adaptable robots while deepening our comprehension of the remarkable feats of animal locomotion.