Self-Evolving AI Robot System for Lunar Exploration and Human Outpost Construction. | Moon Shot Project Goal 3 Realization of AI robots that autonomously learn, adapt to their environment, evolve in intelligence and act alongside human beings, by 2050.

Design, prototyping, and functional analysis of a self-regenerating modular robot

Summary

n this research theme, we designed, fabricated, and analyzed the functions of a modular robot that can adaptively change its form and perform “transformable” tasks, and developed the modular robot system that forms the core of this project. The robot module configuration method, the form of the robot created by combining modules, and the robot behavior and functions (tasks) in each form (design) were each accumulated as a database (repository) and utilized for the simultaneous optimization of structure and controller based on hierarchical reinforcement learning in R&D item 2. The database will be used for simultaneous optimization of structures and controllers based on hierarchical reinforcement learning in R&D item 2.

1.Repository for the Structure and Control of Modular Robots:

We carried out simultaneous optimization of the structure and control of robots in uneven environments, also considering fault tolerance.

2.Design and Prototyping of Connector for Modular Robots:

We determined the functions of individual modules and designed and developed connector modules that include mechanical and electrical coupling mechanisms to enable robotic reconfiguration. Additionally, regarding the informational coupling, we defined a dedicated communication protocol for the modular system. Furthermore, we prototyped an engineering model of a multi-limbed modular robot, MoonBot 0, and validated basic functions such as automatic recognition of the coupled state and autonomous motion generation. We evaluated its control algorithms through operational tests as described later.

3.Development of Module Reconfiguration Algorithms and Implementation in Prototype Robots:

We developed module reconfiguration algorithms to transition from any given structure to another within the database of robot module configurations and implemented these algorithms in prototype robots.

Control of exploration and assembly tasks by multiple robots with different structures

Based on the robot manipulator system, we created scenarios where multiple robots with different structures perform collaborative tasks by restricting the active degrees of freedom of each arm and varying the end effectors. This led to the establishment of a research environment (Sim2Real) that seamlessly integrates dynamic simulations with hardware experiments. We conducted verification on the following items:

(1) For numerous different structures and configurations of the multi-limbed modular robot, we were able to construct control models for each.

(2) Using a robot arm, we demonstrated the capability to collect arbitrary gravel samples from the surface and from up to 10 cm beneath the surface in a simulated gravel field, where rocks with diameters ranging from 1 mm to approximately 50 cm are randomly distributed.

RESEARCH & DEVELOPMENT ITEMS

Design, prototyping, and functional analysis of a self-regenerating modular robot

Functional Analysis Hierarchical Reinforcement Learning for Distributed AI

Integrated control of base construction with multiple robots