From the Lab to the Desert: Fast Prototyping and Learning of Robot Locomotion

Abstract

Abstract-We present a methodology for fast prototyping of morphologies and controllers for robot locomotion. Going beyond simulation-based approaches, we argue that the form and function of a robot, as well as their interplay with realworld environmental conditions are critical. Hence, fast design and learning cycles are necessary to adapt robot shape and behavior to their environment. To this end, we present a combination of laminate robot manufacturing and sampleefficient reinforcement learning. We leverage this methodology to conduct an extensive robot learning experiment. Inspired by locomotion in sea turtles, we design a low-cost crawling robot with variable, interchangeable fins. Learning is performed using both bio-inspired and original fin designs in an artificial indoor environment as well as a natural environment in the Arizona desert. The findings of this study show that static policies developed in the laboratory do not translate to effective locomotion strategies in natural environments. In contrast to that, sample-efficient reinforcement learning can help to rapidly accommodate changes in the environment or the robot.

Computational design of passive grippers

“…However, the revolution in digital fabrication and the resulting ease of customization has opened a new era of task-specific robot design. A body of recent work has shown the advantages of jointly optimizing a robot's shape and actuation for a variety of tasks such as ground locomotion [Digumarti et al 2014;Ha et al 2017;Luck et al 2020;Spielberg et al 2019;, flying [Du et al 2016], swimming [Ma et al 2021], and grasping [Chen et al 2020;Deimel et al 2017;Hazard et al 2018;Pan et al 2020;Xu et al 2021]. Our work builds on this new trend, but instead of customizing a whole robot, we propose to enhance a general-purpose robot with customized end-effectors that can be rapidly fabricated, lowering the cost of customization.…”

Learning to Walk in the Real World with Minimal Human Effort

“…One major challenge is the need to reset the robot to the proper initial states after each episode of data collection, for hundreds or even thousands of roll-outs. Researchers tackle this issue by developing external resetting devices for lightweight robots, such as a simple 1-DoF system [26] or an articulated robotic arm [39]. Otherwise, the learning process requires a large number of manual resets between roll-outs [28,16,54], which limits the scalability of the learning system.…”

“…During training, the robot may fall and damage itself, or leave the training area, which will require labor-intensive human intervention. Because of this, prior work that studied learning locomotion in the real world has focused on statically stable robots [26,39] or relied on tedious manual resets between roll-outs [28,16]. Minimizing human interventions is the key to a scalable reinforcement learning system.…”

Factors Affecting Flipper-Based Robotic Locomotion in Complex Terrains

“…Despite current advancements, there remain notable gaps in the research, particularly concerning the relationship between gait patterns and flipper/body morphology across various challenging terrestrial terrains and the energy efficiency of these systems. Previous works exhibited some limitations, including the exclusive use of front flippers (21,29), reliance on a single terrestrial gait (28), notable gaps in exploring the relationship between gait patterns and flipper/body morphology for various challenging terrestrial terrains, and an analysis of the performance of the robot in only one type of terrestrial terrain. These constraints require a more holistic approach that considers the interplay between gait patterns, flipper morphology, and their implications on robotic design in various environments.…”