Master's Thesis: Developing Methods for Energy-Efficient Gait Patterns in Quadrupedal Robots

Legged robots have remarkable potential to navigate complex terrains and overcome challenging obstacles—yet achieving energy efficiency remains an ongoing challenge. Inspired by the biological concept of embodied intelligence, we have developed the quadrupedal robot Ebert (Fig. 1), designed to harness joint elasticity and finely tuned body resonance to enhance energy efficiency across diverse locomotion speeds.

Nature provides compelling examples of optimized movement patterns. Hoyt and Taylor notably demonstrated how animals instinctively select gait patterns that minimize energy use at different speeds (Fig. 2). Drawing inspiration from these insights, this thesis will explore whether similar adaptive gait behaviors can emerge in robotics through advanced machine learning methods.

You will leverage machine learning algorithms to identify optimal gait patterns for Ebert at varying speeds. Initially, these techniques will be implemented and refined in simulation, followed by validation on real hardware to bridge the gap between theory and application.

Your Qualification:

• Advanced programming skills (Matlab, Python, C++)
• Practical experience with machine learning methods
• Strong background in mathematics and mechanics, particularly non-linear dynamics
• Prior experience with robotic systems and low-level programming is advantageous

Your Research Tasks:

• Develop a gait optimization framework in simulation environments (MuJoCo or Isaac Sim)
• Generate a Hoyt/Taylor energy efficiency curve (Fig. 2) for Ebert
• Validate optimized gaits through hardware experiments

Your Start:

The thesis will be supervised by the Institute of Robotics and Mechatronics at the German Aerospace Center (DLR) in Oberpfaffenhofen, with a starting date at the earliest opportunity.