Autonomous Robots

, 28:151

First online:

Reactive direction control for a mobile robot: a locust-like control of escape direction emerges when a bilateral pair of model locust visual neurons are integrated

  • Shigang YueAffiliated withSchool of Computer Science, University of LincolnBrain Mapping Unit, Sir William Hardy Building, Downing Site, University of CambridgeRidley Building, School of Biology and Psychology, University of Newcastle upon Tyne Email author 
  • , Roger D. SanterAffiliated withSchool of Biological Sciences, University of Nebraska–Lincoln
  • , Yoshifumi YamawakiAffiliated withDepartment of Biology, Faculty of Science, Kyushu University
  • , F. Claire RindAffiliated withRidley Building, School of Biology and Psychology, University of Newcastle upon Tyne

Rent the article at a discount

Rent now

* Final gross prices may vary according to local VAT.

Get Access


Locusts possess a bilateral pair of uniquely identifiable visual neurons that respond vigorously to the image of an approaching object. These neurons are called the lobula giant movement detectors (LGMDs). The locust LGMDs have been extensively studied and this has lead to the development of an LGMD model for use as an artificial collision detector in robotic applications. To date, robots have been equipped with only a single, central artificial LGMD sensor, and this triggers a non-directional stop or rotation when a potentially colliding object is detected. Clearly, for a robot to behave autonomously, it must react differently to stimuli approaching from different directions. In this study, we implement a bilateral pair of LGMD models in Khepera robots equipped with normal and panoramic cameras. We integrate the responses of these LGMD models using methodologies inspired by research on escape direction control in cockroaches. Using ‘randomised winner-take-all’ or ‘steering wheel’ algorithms for LGMD model integration, the Khepera robots could escape an approaching threat in real time and with a similar distribution of escape directions as real locusts. We also found that by optimising these algorithms, we could use them to integrate the left and right DCMD responses of real jumping locusts offline and reproduce the actual escape directions that the locusts took in a particular trial. Our results significantly advance the development of an artificial collision detection and evasion system based on the locust LGMD by allowing it reactive control over robot behaviour. The success of this approach may also indicate some important areas to be pursued in future biological research.


Robots Escape Emergent properties Behaviour Visual neural network LGMD DCMD Locusts Jumping Agents Hybrid Cybernetics