Re-framing bio-plausible collision detection: identifying shared meta-properties through strategic prototyping

Abstract

Insects exhibit remarkable abilities in navigating complex natural environments, whether it be evading predators, capturing prey, or seeking out con-specifics, all of which rely on their compact yet reliable neural systems. We explore the field of bio-inspired robotic vision systems, focusing on the locust inspired Lobula Giant Movement Detector (LGMD) models. The existing LGMD models are thoroughly evaluated, identifying their common meta-properties that are essential for their functionality. This article reveals a common framework, characterized by layered structures and computational strategies, which is crucial for enhancing the capability of bio-inspired models for diverse applications. The result of this analysis is the Strategic Prototype, which embodies the identified meta-properties. It represents a modular and more flexible method for developing more responsive and adaptable robotic visual systems. The perspective highlights the potential of the Strategic Prototype: LGMD-Universally Prototype (LGMD-UP), the key to re-framing LGMD models and advancing our understanding and implementation of bio-inspired visual systems in robotics. It might open up more flexible and adaptable avenues for research and practical applications.

Keywords: LGMD; bio-plausible; environment perception; motion detection; robot; visual sensing.

Figure 1

The LGMD neurons and their computational models. (A) The structure of LGMD1 and LGMD2 neurons in locust, which is adapted from Rind et al. (2016). The figure depicts the abundant dendrite trees with fan-shaped structures in the pre-synaptic regions of both neurons. The scale bars in LGMD1 and LGMD2 are 10 and 5μm, respectively. (B) The responds to a computer-generated visual stimuli of an LGMD1 model, illustrating the middle layer's response of the model, which effectively identified the shape of the looming object. The model is based on Yue and Rind (2006). (C) The responds of the LGMD models and neurons, which is adapted from Dewell and Gabbiani (2018). The standard LGMD neuron response is depicted in the upper, whilst the lower portrays the LGMD model's response identically to the actual neuron. (D) One possible LGMD1 model, which is based on Yue and Rind (2006). This is a standard LGMD1 model with four feed-forward connection layers and one hyper-layer connection layer. (E) One possible LGMD2 model, which is based on Fu and Yue (2015). This is a typical LGMD2 model, similar to LGMD1, with “on” and “off” pathways split from the I/E layer and then reconnected in the S layer.

Figure 2

The structure of LGMD-UP. (A) Attributes of the Graph Network, including Input Edge, Node, and Output Edge, each attribute has different properties. Input Edge contain signals of the form R^m×n, it is then go through the Node with a function F:ℜ^m×n ↦ ℜ^a×b. There are two optional pre-processing parameters and two optional post-processing parameter. Output Edge are the signals of the form R^a×b. (B) The LGMD models described in Graph Network. The Perception, Inhibition, and Global Respond Mechanisms are recursive structures, while other layers are feed-forward structures. (C) The functions in each node. They may differ from each other, but they all correspond to the layers of current LGMD models. For example, the Perception node calculates the differences between two inputs: its last output and the current in node.