Jørgen Christian Larsen

Fault-tolerant gait learning and morphology optimization of a polymorphic walking robot

This paper presents experiments with a morphology-independent, life-long strategy for online learning of locomotion gaits. The experimental platform is a quadruped robot assembled from the LocoKit modular robotic construction kit. The learning strategy applies a stochastic optimization algorithm to optimize eight open parameters of a central pattern generator based gait implementation. We observe that the strategy converges in roughly ten minutes to gaits of similar or higher velocity than a manually designed gait and that the strategy readapts in the event of failed actuators. We also optimize offline the reachable space of a foot based on a reference design but finds that the reality gap hardens the successfully transference to the physical robot. To address this limitation, in future work we plan to study co-learning of morphological and control parameters directly on physical robots.

A robot leg with compliant tarsus and its neural control for efficient and adaptive locomotion on complex terrains

Insects, like dung beetles, show fascinating locomotor abilities. They can use their legs to walk on complex terrains (e.g., rocky and curved surfaces) and to manipulate objects. They also exploit their compliant tarsi, increasing the contact area between the legs and surface, to enhance locomotion, and object manipulation efficiency. Besides these biomechanical components, their neural control allows them to move at a proper frequency with respect to their biomechanical properties and to quickly adapt their movements to deal with environmental changes. Realizing these complex achievements on artificial systems remains a grand challenge. As a step towards this direction, we present here our first prototype of an artificial dung beetle-like leg with compliant tarsus by analyzing real dung beetle legs through

Adaptive strategy for online gait learning evaluated on the polymorphic robotic LocoKit

\mu

Adaptive Combinatorial Neural Control for Robust Locomotion of a Biped Robot

μCT scans. Compliant tarsus was designed according to the so-called fin ray effect. Real robot experiments show that the leg with compliant tarsus can efficiently move on rocky and curved surfaces. We also apply neural control, based on a central pattern generator (CPG) circuit and synaptic plasticity, to autonomously generate a proper moving frequency of the leg. The controller can also adapt the leg movement to deal with environmental changes, like different treadmill speeds, within a few steps.

LocoKit: A Robot Construction Kit for Studying and Developing Functional Morphologies

Modular robots are a kind of robots built from mechatronic modules, which can be assembled in many different ways allowing the modular robot to assume a wide range of morphologies and functions. An important question in modular robotics is to which degree modules should be heterogeneous. In this paper we introduce two contributing factors to heterogeneity namely morphological heterogeneity and sub-functional modularization. Respectively, the ideas are to create modules with significantly different morphologies and to spread sub-functionality across modules. Based on these principles we design and implement the Thor robot and evaluate it by participating in the ICRA Planetary Robotic Contingency Challenge. The Thor robot demonstrates that sub-functional modularity and morphological heterogeneity may increase the versatility of modular robots while reducing the complexity of individual modules, which in the longer term may lead to more affordable modular robots.

Increased performance in a bottom-up designed robot by experimentally guided redesign

This paper presents experiments with a morphology-independent, life-long strategy for online learning of locomotion gaits, performed on a quadruped robot constructed from the LocoKit modular robot. The learning strategy applies a stochastic optimization algorithm to optimize eight open parameters of a central pattern generator based gait implementation. We observe that the strategy converges in roughly ten minutes to gaits of similar or higher velocity than a manually designed gait and that the strategy readapts in the event of failed actuators. In future work we plan to study co-learning of morphological and control parameters directly on the physical robot.

Bio-inspired design and movement generation of dung beetle-like legs

Abstract The task of producing steady, stable and energy efficient locomotion in legged robots with the ability to walk in unknown terrain have for many years been a big challenge in robotics. This work is focusing on how different robots build from the modular robotic system, LocoKit by Larsen et al. [1] , performs compared to animals, and also on the similarities between robots an animals. This work shows, that there in robots exist the same connection between cost of transport and the weight of the robots as is true for animals.

Flexible, fpga-based electronics for modular robots

Humans can perform natural and robust walking behavior. They can even quickly adapt to different situations, like changing their walking speed to synchronize with the speed of a treadmill. Reproducing such complex abilities with artificial bipedal systems is still a difficult problem. To tackle this problem, we present here an adaptive combinatorial neural control circuit consisting of reflex-based and central pattern generator (CPG)-based mechanisms. The reflex-based control mechanism basically generates energy-efficient bipedal locomotion while the CPG-based mechanism with synaptic plasticity ensures robustness against loss of global sensory feedback (e.g., foot contact sensors) as well as allows for adaptation within a few steps to deal with environmental changes. We have successfully applied our control approach to the biomechanical bipedal robot DACBOT. As a result, the robot can robustly walk with energy efficiency and quickly adapt to different speeds of a treadmill.