Shrinking POMCP: A Framework for Real-Time UAV Search and Rescue

TL;DR

A comprehensive approach to optimize UAV-based search and rescue operations in neighborhood areas, utilizing both a 3D AirSim-ROS2 simulator and a 2D simulator, and proposes a novel "Shrinking POMCP" approach to address time constraints.

Abstract

Efficient path optimization for drones in search and rescue operations faces challenges, including limited visibility, time constraints, and complex information gathering in urban environments. We present a comprehensive approach to optimize UAV-based search and rescue operations in neighborhood areas, utilizing both a 3D AirSim-ROS2 simulator and a 2D simulator. The path planning problem is formulated as a partially observable Markov decision process (POMDP), and we propose a novel ``Shrinking POMCP'' approach to address time constraints. In the AirSim environment, we integrate our approach with a probabilistic world model for belief maintenance and a neurosymbolic navigator for obstacle avoidance. The 2D simulator employs surrogate ROS2 nodes with equivalent functionality. We compare trajectories generated by different approaches in the 2D simulator and evaluate performance across various belief types in the 3D AirSim-ROS simulator. Experimental results from both simulators demonstrate that our proposed shrinking POMCP solution achieves significant improvements in search times compared to alternative methods, showcasing its potential for enhancing the efficiency of UAV-assisted search and rescue operations.

Figures (7)

• Figure 1: Our problem features planning the mission of a drone in a neighborhood to search for some targets. The drone is not aware of the real-locations and only have access to the likelihood of targets.
• Figure 2: Four key components of our approach. This paper focuses on the planner component.
• Figure 3: A belief tree constructed by the Shrinking POMCP approach, illustrating its unique decision-making process. Circular nodes represent belief states, with their normalized probability $P(s)$ and quadrotor position $\gamma(s) = (x_q, y_q)$. Black-filled circles indicate non-sparse regions where $P(s) > P_\varepsilon$. Square nodes represent actions. The red arrows show the action sequence $\{a_1, a_2, \ldots, a_k\}$ decided by the agent, where each $a_i = \mathop{\mathrm{arg\,max}}\limits_a Q(b_i, a)$. This sequence terminates upon reaching either a non-sparse region (black node) or the maximum depth $\text{max\_level}$. Unlike standard MCTS, this approach efficiently guides the agent towards high-probability areas, terminating when $P(b_{k+1}) > P_\varepsilon$ or $k = \text{max\_level}$, thus avoiding goal sampling oscillation.
• Figure 4: Environments for different belief scenarios. Left: Uniform belief distribution across the environment. Center: Sparse belief with a single peak, indicating high certainty in one area. Right: Sparse belief with three peaks, representing multiple areas of high certainty. The green triangle ($\blacktriangle$) represents the start position, and the purple star ($\star$) indicates the goal position. Color intensity corresponds to belief value, with darker blue indicating higher belief.
• Figure 5: Comparison between our Shrinking approach and other methods (MCTS, Lawnmower, and Greedy algorithms). Shrinking POMCP requires significantly fewer decision epochs to locate all targets across all belief types. Colors represent different approaches: Blue - Shrinking, Orange - MCTS, Green - Lawnmower, Red - Greedy.