Low-Latency Event-Based Velocimetry for Quadrotor Control in a Narrow Pipe

TL;DR

This work tackles stable quadrotor flight inside confined pipes where self-induced recirculation creates unsteady disturbances. It combines a low-latency event-based smoke velocimetry pipeline with a recurrent disturbance estimator and a reinforcement-learning controller to achieve hovering and controlled lateral translation informed by real-time flow fields. The approach yields sub-millisecond flow processing, real-time disturbance estimates, and notable performance gains, including up to 29% reduction in hover deviation and up to 71% reduction in overshoot during lateral maneuvers. By demonstrating flow-aware closed-loop control in a highly aerodynamically complex environment, the study opens new directions for robust aerial robotics in confined and turbulent settings and provides insights into pipe flow structures and their interaction with small drones.

Abstract

Autonomous quadrotor flight in confined spaces such as pipes and tunnels presents significant challenges due to unsteady, self-induced aerodynamic disturbances. Very recent advances have enabled flight in such conditions, but they either rely on constant motion through the pipe to mitigate airflow recirculation effects or suffer from limited stability during hovering. In this work, we present the first closed-loop control system for quadrotors for hovering in narrow pipes that leverages real-time flow field measurements. We develop a low-latency, event-based smoke velocimetry method that estimates local airflow at high temporal resolution. This flow information is used by a disturbance estimator based on a recurrent convolutional neural network, which infers force and torque disturbances in real time. The estimated disturbances are integrated into a learning-based controller trained via reinforcement learning. The flow-feedback control proves particularly effective during lateral translation maneuvers in the pipe cross-section. There, the real-time disturbance information enables the controller to effectively counteract transient aerodynamic effects, thereby preventing collisions with the pipe wall. To the best of our knowledge, this work represents the first demonstration of an aerial robot with closed-loop control informed by real-time flow field measurements. This opens new directions for research on flight in aerodynamically complex environments. In addition, our work also sheds light on the characteristic flow structures that emerge during flight in narrow, circular pipes, providing new insights at the intersection of robotics and fluid dynamics.

Figures (14)

• Figure 1: In this work, we demonstrate how low-latency event-based smoke velocimetry can be used for real-time disturbance estimation to improve the closed-loop control performance of a quadrotor flying inside a narrow pipe.
• Figure 2: Event frames of the turbulent airflow inside the pipe. The frames are spaced [10]ms apart and polarity-integrated over a [6]ms window. Characteristic smoke structures change appearance on very short timescales, necessitating a high velocimetry update rate and the use of larger spatial patches rather than dense, pixel-level flow estimates.
• Figure 3: Our network architecture used to estimate the aerodynamic disturbances. The ConvLSTM-based flow encoder outputs a latent representation of the flow configuration. This is concatenated with the position of the quadrotor and fed into an MLP to estimate the aerodynamic disturbances. During training we also train with an additive bias which is only a function of the experiment index and accounts for experimental biases, e.g. battery placement.
• Figure 4: Overview of our monocular event-camera motion capture system: the quadrotor is equipped with $N>=4$ infrared LED markers that blink at different frequencies. A single, calibrated event-camera is used to detect all markers and estimate the pose of the object by solving the perspective-n-points problem (PnP) with SqPnP terzakis2020sqpnp.
• Figure 5: Illustration on the construction of the Signed Delta-Time Volume (SDTV) from an event stream. a) The LED is blinking with a period of $\unit[300]{\mu s}$ and a duty cycle of [10]%. b) A single pixel of the event camera records a noisy signal of this blinking LED. False double events (e.g., at $t = \unit[150]{\mu s}, \unit[165]{\mu s}$) and spurious events (e.g., at $t = \unit[630]{\mu s}$) are included. c) Construction of the SDTV illustrated before and after processing the latest time window. d) Periods robustly identified from the SDTV by summing up absolute time differences between negative $\rightarrow$ positive transitions (the first positive value is included). All events until the first positive $\rightarrow$ negative transition are discarded.