Search

This paper explores current developments in evolutionary and bio-inspired approaches to autonomous robotics, concentrating on research from our group at the University of Sussex. These developments are discussed in the context of advances in the wider fields of adaptive and evolutionary approaches to AI and robotics, focusing on the exploitation of embodied dynamics to create behaviour. Four case studies highlight various aspects of such exploitation. The first exploits the dynamical properties of a physical electronic substrate, demonstrating for the first time how component-level analog electronic circuits can be evolved directly in hardware to act as robot controllers. The second develops novel, effective and highly parsimonious navigation methods inspired by the way insects exploit the embodied dynamics of innate behaviours. Combining biological experiments with robotic modeling, it is shown how rapid route learning can be achieved with the aid of navigation-specific visual information that is provided and exploited by the innate behaviours. The third study focuses on the exploitation of neuromechanical chaos in the generation of robust motor behaviours. It is demonstrated how chaotic dynamics can be exploited to power a goal-driven search for desired motor behaviours in embodied systems using a particular control architecture based around neural oscillators. The dynamics are shown to be chaotic at all levels in the system, from the neural to the embodied mechanical. The final study explores the exploitation of the dynamics of brain-body-environment interactions for efficient, agile flapping winged flight. It is shown how a multi-objective evolutionary algorithm can be used to evolved dynamical neural controllers for a simulated flapping wing robot with feathered wings. Results demonstrate robust, stable, agile flight is achieved in the face of random wind gusts by exploiting complex asymmetric dynamics partly enabled by continually changing wing and tail morphologies.

There are several large datasets available that are captured from motion tracking systems which could be useful to train wearable human activity recognition (HAR) systems, if only their spatial data could be mapped into the equivalent inertial measurement unit (IMU) data that would be sensed on the body. In this paper, we describe a mapping from 3D Vicon motion tracking data to data collected from a BlueSense on-body IMU. We characterise the error incurred in order to discern the extent to which it is possible to generate useful training data for a wearable activity recognition system from data collected with a motion capture system. We analyse this by mapping Vicon motion tracking data to rotational velocity and linear acceleration at the head, and compare this to actual gyroscope and accelerometer data collected by an IMU mounted on the head. In a 15 min dataset comprising three static activities—sitting, standing and lying down—we find that 95% of the reconstructed gyroscope data is within an error of [−7.25;+7.46] \(deg \cdot s^{-1}\), while 95% of the reconstructed accelerometer data was contained within [−96.1;+72.9] \(m \cdot G\). However, when we introduce more movement by including data collected while walking this increases to [−19.0;+18.2] \(deg \cdot s^{-1}\) for the gyroscope and [−208;+186] \(m \cdot G\) for the accelerometer. We conclude that generating accurate IMU data from motion capture datasets is possible and could be useful in providing larger volumes of data for activity recognition tasks and in helping enable advanced, data-hungry techniques such as deep learning to be employed on a larger scale within the domain of human activity recognition.

The visual systems of animals have to provide information to guide behaviour and the informational requirements of an animal’s behavioural repertoire are often reflected in its sensory system. For insects, this is often evident in the optical array of the compound eye. One behaviour that insects share with many animals is the use of learnt visual information for navigation. As ants are expert visual navigators it may be that their vision is optimised for navigation. Here we take a computational approach in asking how the details of the optical array influence the informational content of scenes used in simple view matching strategies for orientation. We find that robust orientation is best achieved with low-resolution visual information and a large field of view, similar to the optical properties seen for many ant species. A lower resolution allows for a trade-off between specificity and generalisation for stored views. Additionally, our simulations show that orientation performance increases if different portions of the visual field are considered as discrete visual sensors, each giving an independent directional estimate. This suggests that ants might benefit by processing information from their two eyes independently. Electronic supplementary material The online version of this article (doi:10.1007/s00359-015-1052-1) contains supplementary material, which is available to authorized users.

Drosophila melanogaster are a good system in which to understand the minimal requirements for widespread visually guided behaviours such as navigation, due to their small brains (adults possess only 100,000 neurons) and the availability of neurogenetic techniques which allow the identification of task-specific cell types. Recently published data describe the receptive fields for two classes of visually responsive neurons (R2 and R3/R4d ring neurons in the central complex) that are essential for visual tasks such as orientation memory for salient objects and simple pattern discriminations. What is interesting is that these cells have very large receptive fields and are very small in number, suggesting that each sub-population of cells might be a bottleneck in the processing of visual information for a specific behaviour, as each subset of cells effectively condenses information from approximately 3000 visual receptors in the eye, to fewer than 50 neurons in total. It has recently been shown how R1 ring neurons, which receive input from the same areas as the R2 and R3/R4d cells, are necessary for place learning in Drosophila. However, how R1 neurons enable place learning is unknown. By examining the information provided by different populations of hypothetical visual neurons in simulations of experimental arenas, we show that neurons with ring neuron-like receptive fields are sufficient for defining a location visually. In this way we provide a link between the type of information conveyed by ring neurons and the behaviour they support.

This paper discusses the implementation of insect-inspired visual navigation strategies in flying robots, in particular focusing on the impact of changing height. We start by assessing the information available at different heights for visual homing in natural environments, comparing results from an open environment against one where trees and bushes are closer to the camera. We then test a route following algorithm using a gantry robot and show that a robot would be able to successfully navigate a route at a variety of heights using images saved at a different height.

Cheeseman et al. (1) claim that an ability of honey bees to travel home through a landscape with conflicting information from a celestial compass proves the bees' use of a cognitive map. Their claim involves a curious assumption about the visual information that can be extracted from the terrain: that there is sufficient information for a bee to identify where it is, but insufficient to guide its path without resorting to a cognitive map. We contend that the authors’ claims are unfounded.

Summary Neurogenetic tools of Drosophila research allow unique access to the neural circuitry underpinning visually guided behaviours. New research is highlighting how particular areas in the fly's central brain needed for pattern recognition provide a coarse visual encoding.

The ability of insects to visually navigate long routes to their nest has provided inspiration to engineers seeking to emulate their robust performance with limited resources [1-2]. Many models have been developed based on the elegant snapshot idea: remember what the world looks like from your goal and subsequently move to make your current view more like your memory [3]. In the majority of these models, a single view is stored at a goal location and acts as a form of visual attractor to that position (for review see [4]). Recently however, inspired by the behaviour of ants and the difficulties in extending traditional snapshot models to routes [5], we have proposed a new navigation model [6-7]. In this model, rather than using views to recall directions to the place that they were stored, views are used to recall the direction of facing or movement (identical for a forward-facing ant) at the place the view was stored. To navigate, the agent scans the world by rotating and thus actively finds the most familiar view, a behavior observed in Australian desert ants. Rather than recognise a place, the action to take at that place is specified by a familiar view.