Biologically inspired obstacle avoidance - a technology independent paradigm

Derek Abbott, Andre Yakovleff, Alireza Moini, X. Thong Nguyen, Andrew J. Blanksby, Richard Beare, Andrew Beaumont-Smith, Gyudong Kim, Abdesselam Bouzerdoum, Robert E. Bogner, Kamran EshraghianMobile Robots 1995 · DOI

TL;DR

This paper introduces a biologically inspired, technology-independent smart sensor for obstacle avoidance, mimicking insect vision. This approach offers a multifunctional, compact, and adaptable solution for various applications, moving beyond single-technology, function-specific systems.

What Problem Does This Paper Address?

Traditional approaches to obstacle detection and avoidance in robotics and automotive systems have often been technology-centric, relying on a single sensor modality (e.g., millimeter wave or infrared) and leading to function-specific solutions. This creates systems that are inflexible, quickly become obsolete as technology advances, and are limited in their capabilities. Furthermore, many existing systems use computationally intensive, high-resolution imaging, which is overkill for tasks like collision avoidance and requires bulky hardware. The authors aim to address this by proposing a paradigm shift towards technology-independent, multifunctional systems inspired by the efficient and adaptable visual processing found in insects.

Methods

The core of the proposed solution is a VLSI chip that mimics the insect vision system, specifically focusing on processing optical flow to detect and locate obstacles. This is achieved through a parallel array of photodetectors and analog differentiators that detect contrast changes. The processing at each node is designed to mirror the insect's parallel visual system, enabling improved edge discrimination even in varying light conditions. A key feature is the use of a "Horridge template" which encodes spatio-temporal changes in contrast. This template, combined with a lookup table, allows the system to infer information about object direction, angular velocity, bearing, and time-to-impact.

The technology independence is a crucial aspect: the same sensor concept can be implemented with optical, infrared, or millimeter wave detectors. This allows for easy data fusion if multiple sensor types are used. Multifunctionality is also a significant advantage, as the same chip can perform tasks like braking distance warning, lane tracking, and blind spot detection, unlike conventional function-specific sensors. The paper also proposes Gallium Arsenide (GaAs) as a promising future medium for realizing highly integrated, multifunctional systems due to its high speed, low power consumption, and suitability for integrating various sensor types and communication capabilities.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Technology Independence is Key: The paper strongly advocates for a paradigm shift away from "pet" technologies towards solutions that can adapt to any front-end sensor technology (optical, infrared, millimeter wave). This makes systems more robust to technological evolution.
  • Multifunctionality from Simplicity: By mimicking insect vision, the system produces simple, interpretable "percepts" (like "close," "moving fast," "looming") rather than raw metric data. This simplicity enables a single system to perform multiple tasks, unlike traditional function-specific sensors.
  • Insect Vision as a Model for Efficient Processing: Insects, despite their simple visual systems, perform complex tasks. Their approach, focusing on motion detection and contrast changes rather than full image resolution, offers a blueprint for efficient, low-computation visual processing suitable for real-time applications.
  • GaAs for Future Integration: The paper highlights GaAs as an ideal material for future development, enabling the integration of multiple sensor types (optical, infrared, millimeter wave) and inter-chip communication, paving the way for highly advanced, integrated vision systems.
  • Time-to-Impact Estimation: The system can estimate time-to-impact by analyzing the rate of expansion of an object's visual angle, a biologically relevant cue that can be directly used for collision avoidance.

Key Results

The paper describes the development of a series of "Bugeye" chips, starting with a proof-of-concept in 1992 and evolving through several generations using CMOS and BiCMOS technologies. The latest designs feature 2D arrays of detectors and multiplicative noise cancellation (MNC) for improved performance under varying lighting conditions. The system has demonstrated the ability to extract bearing, velocity, and time-to-impact from visual input. The authors propose that these chips can perform multiple functions, such as blind spot detection and lane sensing, and are significantly more compact and visually discriminating than traditional CCD or discrete infrared approaches.

Limitations and Future Work

The paper acknowledges that current implementations are primarily based on optical sensors. Future work includes developing infrared versions for improved performance in poor weather conditions and exploring the integration of millimeter wave sensors for all-weather capabilities. The authors also propose GaAs as the ideal future platform for realizing highly integrated systems that can support multi-technology fusion and inter-chip communication. A limitation of the current approach, as implied by the discussion on object association, is the potential for tracks to not always correspond to the same object, especially during sensor rotation. Further research is needed to refine these aspects.

Summary updated

BibTeX
@inproceedings{abbott1995biologically,
  author = {Derek Abbott and Andre Yakovleff and Alireza Moini and X. Thong Nguyen and Andrew J. Blanksby and Richard Beare and Andrew Beaumont-Smith and Gyudong Kim and Abdesselam Bouzerdoum and Robert E. Bogner and Kamran Eshraghian},
  title = {{Biologically inspired obstacle avoidance - a technology independent paradigm}},
  booktitle = {Mobile Robots},
  year = {1995},
  doi = {10.1117/12.228964},
}
Figure