Using the Neural Circuit of the Insect
Central Complex for Path Integration on
a Micro Aerial Vehicle
Jan Stankiewicz
(B)
and Barbara Webb
Institute of Perception, Action and Behaviour, School of Informatics,
University of Edinburgh, Edinburgh EH8 9AB, UK
J.stankiewicz@ed.ac.uk
Abstract. We have deployed an anatomically constrained neural model
for path integration on a real world, holonomic aerial platform. Based on
the insect central complex, the model combines estimated heading and
ground speed information to maintain a location estimate that can be
used to steer the agent directly home after convoluted outward journeys.
We implement a biologically plausible method to estimate ground speed
using optical flow. We discover that a downward viewing, mechanically
stabilised and height compensated vision system performs well in a range
of natural environments, even when visual acuity is reduced to 3
◦ /pixel.
In a flat outdoor environment, the worst case final displacement error
increases at a rate 1.5 m 100 m outbound travel. The field of view of the
vision system has no impact on odometry performance.
Keywords: Central complex · Bio-mimetic · MAV · Computer vision
1 Introduction
Many species of flying insect routinely conduct long foraging journeys from which
they can return directly to their starting location. We have recently developed
an anatomically constrained model of how this behaviour can be accomplished
by a path integration (PI) circuit located in the central complex (CX) of the
archetypal insect brain [11]. Conceptually this circuit forms an allothetic navigation scheme that is globally anchored via a sky compass. It has been tested in
simulation and over small distances on an indoor wheeled robot, but as yet has
not been evaluated under the same natural conditions as a flying insect. Here,
we present the results of deploying the CX model on an autonomous micro-aerial
vehicle (MAV).
A key challenge was how to obtain biologically plausible ground speed estimates. Flying insects are thought to sense their egomotion primarily through
Supported by the Edinburgh Centre for Robotics and the Engineering and Physical
Sciences Research Council. We thank Stanley Heinze who provided the neural data
and the initial illustration for Fig. 3. Thanks also to Jiale Lu for his initial work.
c
Springer Nature Switzerland AG 2020
V. Vouloutsi et al. (Eds.): Living Machines 2020, LNAI 12413, pp. 325–337, 2020.
https://doi.org/10.1007/978-3-030-64313-3_31
Central Complex for Path Integration on
a Micro Aerial Vehicle
Jan Stankiewicz
(B)
and Barbara Webb
Institute of Perception, Action and Behaviour, School of Informatics,
University of Edinburgh, Edinburgh EH8 9AB, UK
J.stankiewicz@ed.ac.uk
Abstract. We have deployed an anatomically constrained neural model
for path integration on a real world, holonomic aerial platform. Based on
the insect central complex, the model combines estimated heading and
ground speed information to maintain a location estimate that can be
used to steer the agent directly home after convoluted outward journeys.
We implement a biologically plausible method to estimate ground speed
using optical flow. We discover that a downward viewing, mechanically
stabilised and height compensated vision system performs well in a range
of natural environments, even when visual acuity is reduced to 3
◦ /pixel.
In a flat outdoor environment, the worst case final displacement error
increases at a rate 1.5 m 100 m outbound travel. The field of view of the
vision system has no impact on odometry performance.
Keywords: Central complex · Bio-mimetic · MAV · Computer vision
1 Introduction
Many species of flying insect routinely conduct long foraging journeys from which
they can return directly to their starting location. We have recently developed
an anatomically constrained model of how this behaviour can be accomplished
by a path integration (PI) circuit located in the central complex (CX) of the
archetypal insect brain [11]. Conceptually this circuit forms an allothetic navigation scheme that is globally anchored via a sky compass. It has been tested in
simulation and over small distances on an indoor wheeled robot, but as yet has
not been evaluated under the same natural conditions as a flying insect. Here,
we present the results of deploying the CX model on an autonomous micro-aerial
vehicle (MAV).
A key challenge was how to obtain biologically plausible ground speed estimates. Flying insects are thought to sense their egomotion primarily through
Supported by the Edinburgh Centre for Robotics and the Engineering and Physical
Sciences Research Council. We thank Stanley Heinze who provided the neural data
and the initial illustration for Fig. 3. Thanks also to Jiale Lu for his initial work.
c
Springer Nature Switzerland AG 2020
V. Vouloutsi et al. (Eds.): Living Machines 2020, LNAI 12413, pp. 325–337, 2020.
https://doi.org/10.1007/978-3-030-64313-3_31