Olympus: A Jumping Quadruped for Planetary Exploration Utilizing Reinforcement Learning for In-Flight Attitude Control

TL;DR

Olympus addresses planetary exploration under low gravity by integrating a Mars-optimized jumping quadruped with a 5-bar leg design and a PPO-based in-flight attitude controller. The system design is coupled with a high-fidelity sim2real strategy, including motor-model identification and domain randomization, and is validated through rod and rope experiments alongside simulation studies. Key contributions include a design-optimization workflow for leg/body parameters, a learning-based controller for stabilizing attitude during flight, and demonstrable sim2real transfer in physical tests. The work has practical significance for robust, agile exploration of challenging terrains like Martian lava tubes, enabling larger obstacles to be surmounted via jumping while maintaining safe landings and reorientation capabilities.

Abstract

Exploring planetary bodies with lower gravity, such as the moon and Mars, allows legged robots to utilize jumping as an efficient form of locomotion thus giving them a valuable advantage over traditional rovers for exploration. Motivated by this fact, this paper presents the design, simulation, and learning-based "in-flight" attitude control of Olympus, a jumping legged robot tailored to the gravity of Mars. First, the design requirements are outlined followed by detailing how simulation enabled optimizing the robot's design - from its legs to the overall configuration - towards high vertical jumping, forward jumping distance, and in-flight attitude reorientation. Subsequently, the reinforcement learning policy used to track desired in-flight attitude maneuvers is presented. Successfully crossing the sim2real gap, extensive experimental studies of attitude reorientation tests are demonstrated.

Figures (6)

• Figure 1: The Olympus quadruped mounted for experimental setups for reorientation tests - roll evaluation on rotating rod (left), pitch assessment on rotating rod (center), and yaw test using fixed rope suspension (right).
• Figure 2: CAD render of Olympus, depicting a) the core component lengths of the robot, b) highlight of the $3$ motors of each leg for the front right motor (yellow used for the 5-bar motors and cyan for the hip motor), as well as c) where the springs are integrated within each leg.
• Figure 3: Parameter space exploration of body dimensions: body length $l_{body}$, front leg separation $w_{body f}$, and back leg separation $w_{body f}$. Heat map illustrates performance correlation with body size parameters for roll, pitch, and yaw reorientation in the grid search optimization, maintaining fixed link lengths.
• Figure 4: Reinforcement learning controller architecture employed for the problem of in-flight attitude control of the Olympus quadruped.
• Figure 5: Closed-loop performance of the RL policy to step inputs in either roll, pitch, or yaw for different setup configurations.