Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bio-Inspired Event-Based Visual Servoing for Ground Robots

Maral Mordad, Kian Behzad, Debojyoti Biswas, Noah J. Cowan, Milad Siami

Abstract

Biological sensory systems are inherently adaptive, filtering out constant stimuli and prioritizing relative changes, likely enhancing computational and metabolic efficiency. Inspired by active sensing behaviors across a wide range of animals, this paper presents a novel event-based visual servoing framework for ground robots. Utilizing a Dynamic Vision Sensor (DVS), we demonstrate that by applying a fixed spatial kernel to the asynchronous event stream generated from structured logarithmic intensity-change patterns, the resulting net event flux analytically isolates specific kinematic states. We establish a generalized theoretical bound for this event rate estimator and show that linear and quadratic spatial profiles isolate the robot's velocity and position-velocity product, respectively. Leveraging these properties, we employ a multi-pattern stimulus to directly synthesize a nonlinear state-feedback term entirely without traditional state estimation. To overcome the inescapable loss of linear observability at equilibrium inherent in event sensing, we propose a bio-inspired active sensing limit-cycle controller. Experimental validation on a 1/10-scale autonomous ground vehicle confirms the efficacy, extreme low-latency, and computational efficiency of the proposed direct-sensing approach.

Bio-Inspired Event-Based Visual Servoing for Ground Robots

Abstract

Biological sensory systems are inherently adaptive, filtering out constant stimuli and prioritizing relative changes, likely enhancing computational and metabolic efficiency. Inspired by active sensing behaviors across a wide range of animals, this paper presents a novel event-based visual servoing framework for ground robots. Utilizing a Dynamic Vision Sensor (DVS), we demonstrate that by applying a fixed spatial kernel to the asynchronous event stream generated from structured logarithmic intensity-change patterns, the resulting net event flux analytically isolates specific kinematic states. We establish a generalized theoretical bound for this event rate estimator and show that linear and quadratic spatial profiles isolate the robot's velocity and position-velocity product, respectively. Leveraging these properties, we employ a multi-pattern stimulus to directly synthesize a nonlinear state-feedback term entirely without traditional state estimation. To overcome the inescapable loss of linear observability at equilibrium inherent in event sensing, we propose a bio-inspired active sensing limit-cycle controller. Experimental validation on a 1/10-scale autonomous ground vehicle confirms the efficacy, extreme low-latency, and computational efficiency of the proposed direct-sensing approach.
Paper Structure (12 sections, 2 theorems, 35 equations, 4 figures, 1 table)

This paper contains 12 sections, 2 theorems, 35 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Problem Formulation and Modeling
  3. Event Camera Model and Event Processing
  4. Longitudinal Dynamics of the Robot
  5. Synthesis of Event-Driven Feedback
  6. Active Sensing Strategy
  7. Experiments and Simulations
  8. Motion Dynamics Parameters
  9. Net Event Count Estimator Results
  10. EBVS with Active Sensing
  11. Conclusion and Future Work
  12. Acknowledgment

Key Result

Theorem 1

Let $x(t) \in C^2(\mathbb{R})$ denote the camera position with instantaneous linear velocity $\dot{x}(t)$ and acceleration $\ddot{x}(t)$. Assume the camera observes a scene with an arbitrary horizontally exponential intensity profile $I_0(u, v) = \exp(f(u)), \: \forall v$, where $f \in C^3(\mathbb{R where the net event rate estimator $M$ is evaluated at the image-plane coordinate $\mu(t)=-\frac{f_

Figures (4)

  • Figure 1: Experimental setup and control architecture.(A) A ground vehicle moves parallel to a monitor displaying quadratic and linear intensity patterns for an event camera. RGB axes denote the $x, y, z$ coordinate frames for the robot, camera, and world (origin at the pattern's stabilization target). (B) Closed-loop EBVS with active sensing: Net event counts from the stimuli regulate vehicle-driven camera motion to stabilize it at the desired center.
  • Figure 2: (A) Dual-pattern display with quadratic (top) and linear (bottom) intensity profiles. (B) Corresponding accumulated event stream, where white and blue dots indicate positive and negative polarity events, respectively. The robot's motion to the right relative to the pattern generates the observed events. Red rectangles denote the kernels $\mathcal{K}_1$ and $\mathcal{K}_2$.
  • Figure 3: Event-based state estimation vs. ground truth. Top: $x(t)\dot{x}(t)$ from $\mathcal{K}_1$. Bottom: $\dot{x}(t)$ from $\mathcal{K}_2$. Shaded regions indicate theoretical bounds. Ground truth for the robot's position is provided by an array of six overhead motion capture cameras, while ground truth velocity is measured directly via the robot's onboard wheel encoders.
  • Figure 4: Closed-loop EBVS response for (A) fixed and (B) time-varying stabilization points. In both cases, in the left panel, the top plot shows position and the bottom plot shows velocity, with the desired oscillation radius $a$ indicated by dashed black lines. The right panels present the phase portraits, where green and red markers denote the start and end of the trajectory, respectively.

Theorems & Definitions (7)

  • Definition 1: Net Event Count
  • Theorem 1
  • proof
  • Example 1: Quadratic Profile
  • Example 2: Linear Profile
  • Theorem 2: Extension from Biswas et al. biswas2025exact
  • proof