Bio-Inspired Event-Based Visual Servoing for Ground Robots

Area of Science:

• Robotics
• Computer Vision
• Bio-inspired Engineering

Background:

• Biological systems efficiently process sensory information by focusing on changes.
• Robotic systems often rely on traditional state estimation, which can be computationally intensive.
• Event-based sensing offers potential for low-latency and efficient perception.

Purpose of the Study:

• To develop a principled 1D event-based visual servoing framework for ground robots.
• To leverage Dynamic Vision Sensor (DVS) event streams for direct kinematic state estimation.
• To introduce a bio-inspired controller for enhanced robotic sensing and control.

Main Methods:

• Utilizing a Dynamic Vision Sensor (DVS) to capture asynchronous event streams.
• Applying spatial kernels to event data to isolate kinematic states (velocity, position-velocity).
• Developing a direct nonlinear state feedback synthesis and an active sensing limit-cycle controller.

Main Results:

• Demonstrated analytical isolation of robot velocity and position-velocity product using specific spatial profiles.
• Achieved direct synthesis of nonlinear state feedback without traditional estimation.
• Validated the framework's efficacy, low-latency, and computational efficiency on a 1/10-scale robot.

Conclusions:

• The proposed event-based visual servoing framework offers a computationally efficient and low-latency alternative for robotic control.
• Bio-inspired active sensing strategies can overcome limitations of event-based sensing.
• Direct sensing approaches using DVS show significant promise for autonomous systems.