Confronting the challenge of learning a flexible neural controller for a diversity of morphologies

  title={Confronting the challenge of learning a flexible neural controller for a diversity of morphologies},
  author={Sebastian Risi and Kenneth O. Stanley},
  booktitle={Annual Conference on Genetic and Evolutionary Computation},
The ambulatory capabilities of legged robots offer the potential for access to dangerous and uneven terrain without a risk to human life. However, while machine learning has proven effective at training such robots to walk, a significant limitation of such approaches is that controllers trained for a specific robot are likely to fail when transferred to a robot with a slightly different morphology. This paper confronts this challenge with a novel strategy: Instead of training a controller for a… 

Figures and Tables from this paper

Evolving morphological robustness for collective robotics

  • R. PutterG. Nitschke
  • Engineering
    2017 IEEE Symposium Series on Computational Intelligence (SSCI)
  • 2017
Results indicate that an indirect (developmental) encoding NE method consistently evolves controllers that fully function when transferred to teams with varied morphologies.

Evolving a Spiking Neural Network controller for low gravity environments

This thesis addresses the development of a lightweight and quickly trainable controller based on bio-inspired and time-dependent spiking neural networks, tasked to enact an effective gait for a

Evolving Robust Robot Team Morphologies for Collective Construction

Results indicated that the developmental method was appropriate for evolving controllers that were able to generalize to a range of team morphologies that solved the collective construction task with a high degree of task performance.

Neuro-evolution behavior transfer for collective behavior tasks

Experimental results indicate a hybrid of objective-based search and behavioral diversity maintenance in evolutionary controller design coupled with behavior transfer yields evolved behaviors of significantly high quality across increasingly complex multi-agent tasks.

Objective versus Non-Objective Search in Evolving Morphologically Robust Robot Controllers

  • R. PutterG. Nitschke
  • Biology, Psychology
    2018 IEEE Symposium Series on Computational Intelligence (SSCI)
  • 2018
Behavioral diversity methods such as novelty search mat not be suitable for generating robot behaviors that can continue functioning given changing robot morphologies, for example, due to damaged or disabled sensors and actuators.

Evolving and merging hebbian learning rules: increasing generalization by decreasing the number of rules

It is shown that by allowing multiple connections in the network to share the same local learning rule, it is possible to drastically reduce the number of trainable parameters, while obtaining a more robust agent.

Quality Diversity: A New Frontier for Evolutionary Computation

A new approach is investigated that hybridizes multiple views of behaviors (called behavior characterizations) in the same run, which succeeds in overcoming some of the challenges associated with searching for QD with respect to a behavior characterization that is not necessarily sufficient for generating both quality and diversity at the same time.

Accessible Survey of Evolutionary Robotics and Potential Future Research Directions

The primary objective of this study is to explore the applicability of Evolutionary Approaches in robotic application development and present a brief summary of the popular techniques in adaptive robotics such as GA, Neural Network, NeuroEvolution of Augmenting Topologies and Generative Representation in ER.

Automated Reinforcement Learning (AutoRL): A Survey and Open Problems

This survey seeks to unify the field of AutoRL, provide a common taxonomy, discuss each area in detail and pose open problems of interest to researchers going forward.

Training a Multiagent Hive Brain for Coordinated UGV Operations (4.4.2)

This project builds on a learning method developed previously under DARPA support called multiagent HyperNEAT that evolves a set of neural controllers for a team of collaborating wheeled robots to add directionality to the communication system so that agents can efficiently decide and perceive from where in space signals originate.



Evolving coordinated quadruped gaits with the HyperNEAT generative encoding

It is demonstrated that HyperNEAT, a new and promising generative encoding for evolving neural networks, can evolve quadruped gaits without an engineer manually decomposing the problem.

A Robot that Walks; Emergent Behaviors from a Carefully Evolved Network

  • R. Brooks
  • Computer Science
    Neural Computation
  • 1989
This paper suggests one possible mechanism for analagous robot evolution by describing a carefully designed series of networks, each one being a strict augmentation of the previous one, which control a six-legged walking machine capable of walking over rough terrain and following a person passively sensed in the infrared spectrum.

Evolving the Walking Behaviour of a 12 DOF Quadruped Using a Distributed Neural Architecture

The main target of this paper is to show that the distributed architecture for the general control of robots can be applied to complex dynamic tasks like walking and show how a completely neural and distributed controller can be obtained for a robot ascomplex as Aibo on a task as complex as walking.

GasNets and other Evolvable Neural Networks applied to Bipedal Locomotion

Of the 14 distinct networks tested, GasNets were the only network to achieve cyclical locomotion, although CTRNNs were able to attain a higher level of average fitness.

A Case Study on the Critical Role of Geometric Regularity in Machine Learning

It is argued that geometric information is critical to the ability of any machine learning approach to effectively generalize; even a small shift in the configuration of the task in space from what was experienced in training can go wholly unrecognized unless the algorithm is able to learn the regularities in decision-making across the problem geometry.

GasNets and other evovalble neural networks applied to bipedal locomotion

Of the 14 distinct networks tested, GasNets were the only network to achieve cyclical locomotion, although CTRNNs were able to attain a higher level of average fitness.

Autonomous Evolution of Topographic Regularities in Artificial Neural Networks

This letter shows that when geometry is introduced to evolved ANNs through the hypercube-based neuroevolution of augmenting topologies algorithm, they begin to acquire characteristics that indeed are reminiscent of biological brains.

Neuroevolution: from architectures to learning

This paper gives an overview of the most prominent methods for evolving ANNs with a special focus on recent advances in the synthesis of learning architectures.

Evolution of central pattern generators for bipedal walking in a real-time physics environment

Using a full rigid-body simulation of a biped, it was possible to evolve recurrent neural networks that controlled stable straight-line walking on a planar surface and simple sensory input to locate a sound source was integrated to achieve directional walking.

Guided Optimization for Balanced Locomotion

A method of using guiding forces to allow progressive learning of control actions for balanced locomotion is presented and involves using a “hand of God” to facilitate balance while the basic actions of a desired motion are learned.