Perception, control and planning for (self-)reconfigurable modular robots

Modular robots (MRs) consist of similar modules that can be configured into different shapes. MRs introduce a number of benefits over conventional robots specifically designed for a task. Self-reconfigurable modular robots (SRMRs) are a sub-category of MRs that can perform complex morphological changes by making use of autonomous attachments and detachments. This thesis is pointing out various aspects of MRs with particular interest in SRMRs (but not exclusively). First of all, designs of SRMRs are studied and improvements have been demonstrated in Roombots (RB), a SRMR designed in the Biorobotics Laboratory. The design improvements involve both electromechanical hardware, control and planning aspects. After all improvements, complex scenarios can be studied involving more than 10 modules adding up more than 30 DOF system. Such SRMRs are not intuitive for humans to operate. To overcome this issue, three different user interfaces (UIs) are proposed covering different aspects to simplify use of SRMRs. The first UI is used for conveying shape information to SRMRs using a tangible interaction interface as a proxy. The second one does not have any interaction medium and user directly control SRMRs via gestures that are tracked by external vision systems. The last UI shows possibility of assembling SRMRs in a virtual reality environment using a head mounted display and a hand gesture tracking system. All three of proposed UIs are novel and first of their kind for SRMRs. The final focus is locomotion with modular morphologies and presented in three major studies. Initially, one of the most simplest modular structures, a parallel chain structure, is studied to develop a goal driven neuro-visual locomotion model of a swimming lamprey robot. It is followed by a slightly more complex chain structure. A snake robot is simulated during different serpentine gaits to evaluate implications of different on-board camera placement strategies for such robots. In the last phase, the importance of compliance in legged locomotion is experimentally assessed for quadrupedal modular structures. Compliance of legs as well as a control scheme that acts as a compliant element is taken into consideration for a comparative evaluation. Locomotion is a fundamental part of all mobile robots. Three biologically inspired methods are offered to tackle locomotion of MRs that present performance improvements on multiple aspects as well as better understanding of certain concepts. Although the idea of modularity is widely used in industrial production and common daily items, MRs still don't find much use in daily lives. This thesis brings MRs one step closer to wide daily use by addressing such systems from multiple perspectives.

Control of Bio-Inspired Sprawling Posture Quadruped Robots with an Actuated Spine

Tomislav Horvat

Sprawling posture robots are characterized by upper limb segments protruding horizontally from the body, resulting in lower body height and wider support on the ground. Combined with an actuated segmented spine and tail, such morphology resembles that of salamanders or crocodiles. Although bio-inspired salamander-like robots with simple rotational limbs have been created, not much research has been done on kinematically redundant bio-mimetic robots that can closely replicate kinematics of sprawling animal gaits. Being bio-mimetic could allow a robot to have some of the locomotion skills observed in those animals, expanding its potential applications in challenging scenarios. At the same time, the robot could be used to answer questions about the animal's locomotion. This thesis is focused on developing locomotion controllers for such robots. Due to their high number of degrees of freedom (DoF), the control is based on solving the limb and spine inverse kinematics to properly coordinate different body parts. It is demonstrated how active use of a spine improves the robot's walking and turning performance. Further performance improvement across a variety of gaits is achieved by using model predictive control (MPC) methods to dictate the motion of the robot's center of mass (CoM). The locomotion controller is reused on an another robot (OroBOT) with similar morphology, designed to mimic the kinematics of a fossil belonging to Orobates, an extinct early tetrapod. Being capable of generating different gaits and quantitatively measuring their characteristics, OroBOT was used to find the most probable way the animal moved. This is useful because understanding locomotion of extinct vertebrates helps to conceptualize major transitions in their evolution. To tackle field applications, e.g. in disaster response missions, a new generation of field-oriented sprawling posture robots was built. The robustness of their initial crocodile-inspired design was tested in the animal's natural habitat (Uganda, Africa) and subsequently enhanced with additional sensors, cameras and computer. The improvements to the software framework involved a smartphone user interface visualizing the robot's state and camera feed to improve the ease of use for the operator. Using force sensors, the locomotion controller is expanded with a set of reflex control modules. It is demonstrated how these modules improve the robot's performance on rough and unstructured terrain. The robot's design and its low profile allow it to traverse low passages. To also tackle narrow passages like pipes, an unconventional crawling gait is explored. While using it, the robot lies on the ground and pushes against the pipe walls to move the body. To achieve such a task, several new control and estimation modules were developed. By exploring these problems, this thesis illustrates fruitful interactions that can take place between robotics, biology and paleontology.

EPFL2018

Compliance, locomotion and local computation in (self-) reconfigurable modular robots

Simon Lukas Hauser

Animals display an enormous versatility and a remarkable ability to adapt to changes in environment and terrain. Research in bio-inspired robotics strives to transfer these skills to robots, including legged systems. Even though animals seemingly effortlessly perform most of their activities on rough terrain, this feature seems to be particularly difficult to achieve in their robotic counterparts. We hypothesize that perturbations caused by e.g. rough terrain are not handled by the central nervous system, but rather by local modifications of the locomotion system such as legs and feet. Modifications can include changing mechanical properties such as spring and damping characteristics, as well as adjustments in the movement pattern of the locomotion. In this thesis, we isolate instances where a specific component, in our view, needs such modifications to fulfill different functionalities in a locomotion cycle. Further, we introduce the concept of mode-switches: a local computation is performed to induce these changes in functionalities by changing the dynamical response of the component. We then present one mode-switch method in hardware, jamming of granular media, that can switch the functionality of a foot between impact damping'' and propulsion force transmission'', and we show how this regularizes step sizes on rough terrain. We then present one mode-switch method in control, a force feedback strategy named tegotae''. This switches the functionality of leg movement between displacing the leg'' and

displacing the body'', and we show how this informs the controller about which legs are bearing less weight and thus are more suited to be moved. We suggest that these methods can be applied to any legged structure and use modular robots to demonstrate these concepts. In parallel, we also improved our previously developed self-reconfigurable modular robot platform

Roombots'' such that they perform a variety of tasks centered around adaptive and assistive furniture with up to 12 modules. This includes demonstrations of self-reconfiguration, mobile furniture, object manipulation, interaction capabilities and the development of a user interface. With these improvements, this platform can in the future also be used for further locomotion research where the shape-shifting ability could be of major importance.

EPFL2019

Learning to move in modular robots using central pattern generators and online optimization

Auke Ijspeert, Jérôme Maye

This article addresses the problem of how modular robotics systems, i.e. systems composed of multiple modules that can be configured into different robotic structures, can learn to locomote. In particular, we tackle the problems of online learning, that is, learning while moving, and the problem of dealing with unknown arbitrary robotic structures. We propose a framework for learning locomotion controllers based on two components: a central pattern generator (CPG) and a gradient-free optimization algorithm: Powell’s method. The CPG is implemented as a system of coupled nonlinear oscillators in our YaMoR modular robotic system, with one oscillator per module. The nonlinear oscillators are coupled together across modules using Bluetooth communication to obtain specific gaits, i.e. synchronized patterns of oscillations among modules. Online learning is done by running the Powell optimization algorithm in parallel to the CPG model, with the speed of locomotion being the criterion to be optimized. Interesting aspects of the optimization are: it is carried out online, it does not require stopping or resetting the robots, and it is fast. We present results showing the interesting properties of this framework for a modular robotic system. In particular, our CPG model can readily be implemented in a distributed system, it is cheap computationally, it exhibits limit cycle behavior (temporary perturbations are rapidly forgotten), it produces smooth trajectories even when control parameters are abruptly changed, and it is robust against imperfect communication among modules. We also present results of learning to move with three different robot structures. Interesting locomotion modes are obtained after running the optimization for less than 60 minutes.

2008