Deep Reinforcement Learning based Aggressive Collision Avoidance with Limited FOV for Unmanned Aerial Vehicles

Hasanzade, Mehmet; Shadeed, Omar; Koyuncu, Emre

doi:10.2514/6.2022-2043

Cited by 3 publications

(2 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…An existing issue for UAVs is that the small field of view (FoV) of these sensors severely limits the UAV's perception capability and task efficiency. Although many efforts have been made to deal with the constraints in applications induced by narrow or limited FoVs (1)(2)(3)(4)(5), a larger FoV is still a better solution that not only reduces task time by observing the environment more efficiently (6) but also enhances the UAV safety in the wild by perceiving dynamic obstacles (e.g., birds) approaching from an unknown direction (7).…”

Section: Introductionmentioning

confidence: 99%

A self-rotating, single-actuated UAV with extended sensor field of view for autonomous navigation

et al. 2023

View full text Add to dashboard Cite

Uncrewed aerial vehicles (UAVs) rely heavily on visual sensors to perceive obstacles and explore environments. Current UAVs are limited in both perception capability and task efficiency because of a small sensor field of view (FoV). One solution could be to leverage self-rotation in UAVs to extend the sensor FoV without consuming extra power. This natural mechanism, induced by the counter-torque of the UAV motor, has rarely been exploited by existing autonomous UAVs because of the difficulties in design and control due to highly coupled and nonlinear dynamics and the challenges in navigation brought by the high-rate self-rotation. Here, we present powered-flying ultra-underactuated LiDAR (light detection and ranging) sensing aerial robot (PULSAR), an agile and self-rotating UAV whose three-dimensional position is fully controlled by actuating only one motor to obtain the required thrust and moment. The use of a single actuator effectively reduces the energy loss in powered flights. Consequently, PULSAR consumes 26.7% less power than the benchmarked quadrotor with the same total propeller disk area and avionic payloads while retaining a good level of agility. Augmented by an onboard LiDAR sensor, PULSAR can perform autonomous navigation in unknown environments and detect both static and dynamic obstacles in panoramic views without any external instruments. We report the experiments of PULSAR in environment exploration and multidirectional dynamic obstacle avoidance with the extended FoV via self-rotation, which could lead to increased perception capability, task efficiency, and flight safety.

show abstract

Section: Introductionmentioning

confidence: 99%

A self-rotating, single-actuated UAV with extended sensor field of view for autonomous navigation

et al. 2023

View full text Add to dashboard Cite

show abstract

“…The smoothness of the flight trajectories, however, is not directly addressed. Hasanzade et al [15] proposed an RL-based UAV navigation solution based on a trajectory re-planning algorithm, where high order B-splines are used to define and specify flight trajectories. Due to the local support property of B-spline, such trajectories can be updated quickly, allowing the small UAVs to navigate in clutter environments aggressively.…”

Section: Introductionmentioning

confidence: 99%

Smooth Trajectory Collision Avoidance through Deep Reinforcement Learning

Song¹,

Saunders²,

Yue³

et al. 2022

Preprint

View full text Add to dashboard Cite

Collision avoidance is a crucial task in vision-guided autonomous navigation. Solutions based on deep reinforcement learning (DRL) has become increasingly popular. In this work, we proposed several novel agent state and reward function designs to tackle two critical issues in DRL-based navigation solutions: 1) smoothness of the trained flight trajectories; and 2) model generalization to handle unseen environments.Formulated under a DRL framework, our model relies on margin reward and smoothness constraints to ensure UAVs fly smoothly while greatly reducing the chance of collision. The proposed smoothness reward minimizes a combination of firstorder and second-order derivatives of flight trajectories, which can also drive the points to be evenly distributed, leading to stable flight speed. To enhance the agent's capability of handling new unseen environments, two practical setups are proposed to improve the invariance of both the state and reward function when deploying in different scenes. Experiments demonstrate the effectiveness of our overall design and individual components.

show abstract