Core Concepts
Parallel wire-driven monopedal robot RAMIEL achieves stable continuous jumping through reinforcement learning.
Abstract
RAMIEL, a parallel wire-driven monopedal robot, exhibits high-speed and high-power linear motion capabilities. Despite its performance in jumping, control instability hinders continuous jumps. This study proposes using reinforcement learning to achieve stable continuous jumping motions by inferring joint velocities from joint angles. The simulation results demonstrate the effectiveness of this method for both simulation and actual robot applications.
I. INTRODUCTION
Legged robots excel in traversing uneven terrains by jumping.
RAMIEL's design allows for high and continuous jumps.
II. PARALLEL WIRE-DRIVEN MONOPEDAL ROBOT RAMIEL
Detailed body structure and performance of continuous jumping.
III. CONTINUOUS JUMPING MOTION OF RAMIEL USING REINFORCEMENT LEARNING
System architecture for simulation and actual robot application.
IV. EXPERIMENTS
Simulation experiments show improved performance with reinforcement learning methods.
Actual robot experiments demonstrate successful continuous jumps using reinforcement learning techniques.
V. DISCUSSION
Challenges faced in achieving complete continuous jumping motion on the actual robot are discussed.
VI. CONCLUSION
Reinforcement learning enables dynamic motions in parallel wire-driven robots like RAMIEL.
Stats
"RAMIEL has succeeded in a high jump of 1.6 m and a maximum of eight consecutive jumps."
"10 out of 16 trials resulted in less than two consecutive jumps."
"The motor is driven by a motor driver that can operate at an input voltage of 70V and a maximum instantaneous current of 50A."