Abstract:The design of gaits for robot locomotion can be a daunting process which requires significant expert knowledge and engineering. This process is even more challenging for robots that do not have an accurate physical model, such as compliant or micro-scale robots. Data-driven gait optimization provides an automated alternative to analytical gait design. In this paper, we propose a novel approach to efficiently learn a wide range of locomotion tasks with walking robots. This approach formalizes locomotion as a contextual policy search task to collect data, and subsequently uses that data to learn multi-objective locomotion primitives that can be used for planning. As a proof-of-concept we consider a simulated hexapod modeled after a recently developed microrobot, and we thoroughly evaluate the performance of this microrobot on different tasks and gaits. Our results validate the proposed controller and learning scheme on single and multi-objective locomotion tasks. Moreover, the experimental simulations show that without any prior knowledge about the robot used (e.g., dynamics model), our approach is capable of learning locomotion primitives within 250 trials and subsequently using them to successfully navigate through a maze.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the challenges in the gait design and optimization of micro - robots. Specifically: 1. **Lack of accurate dynamic models**: For micron - scale robots, due to their extremely small size, it is very difficult to obtain an accurate dynamic model. This makes traditional gait design methods difficult to apply. 2. **High - cost experimental iteration**: The experimental iteration of micro - robots is very time - consuming and expensive. Therefore, a method that can quickly learn effective gaits within a limited number of trials is required. 3. **Complex multi - objective optimization**: In practical applications, gait design needs to consider not only walking speed but also multiple objectives such as energy efficiency. How to find the optimal solution among these conflicting objectives is a challenge. To address these challenges, the author proposes a new method to automatically optimize gait in a data - driven manner. This method uses a central pattern generator (CPG) controller combined with Bayesian optimization (BO) technology, and can efficiently learn the gaits and motion primitives required for various motion tasks without prior knowledge. The specific steps are as follows: - **Data collection**: Collect data of different motion primitives through contextual policy search. - **Multi - objective optimization**: Use the collected data for multi - objective optimization to generate gaits that can meet multiple objectives. - **Path planning**: The learned motion primitives can be used for path planning, enabling the robot to successfully navigate complex environments such as mazes. Through these methods, the author verifies the effectiveness and robustness of this method in single - objective and multi - objective motion tasks. The experimental results show that this method can learn effective gaits within 250 trials and be successfully applied to path - planning tasks.

Learning Flexible and Reusable Locomotion Primitives for a Microrobot

Learning Gait-conditioned Bipedal Locomotion with Motor Adaptation

Adaptive Locomotion Control of a Hexapod Robot Via Bio-Inspired Learning

Gait Learning for Soft Microrobots Controlled by Light Fields

Deep Reinforcement Learning Based Co-Optimization of Morphology and Gait for Small-Scale Legged Robot

Contact-Implicit Optimization of Locomotion Trajectories for a Quadrupedal Microrobot

Environment-Centric Learning Approach for Gait Synthesis in Terrestrial Soft Robots

Learning and Adapting Agile Locomotion Skills by Transferring Experience

Learning to Exploit Elastic Actuators for Quadruped Locomotion

Learning Robust, Agile, Natural Legged Locomotion Skills in the Wild

Data-efficient Learning of Morphology and Controller for a Microrobot

Skill Latent Space Based Multigait Learning for a Legged Robot

Effective Locomotion at Multiple Stride Frequencies Using Proprioceptive Feedback on a Legged Microrobot

Experience-Learning Inspired Two-Step Reward Method for Efficient Legged Locomotion Learning Towards Natural and Robust Gaits

CPG-Based Hierarchical Locomotion Control for Modular Quadrupedal Robots Using Deep Reinforcement Learning.

Learning Agile, Robust Locomotion Skills for Quadruped Robot.

Learning plastic matching of robot dynamics in closed-loop central pattern generators

Adaptive Gait Modeling and Optimization for Principally Kinematic Systems

Walking motion generation, synthesis, and control for biped robot by using PGRL, LPI, and fuzzy logic

Smart Gait: A Gait Optimization Framework for Hexapod Robots

Residual Model Learning for Microrobot Control