LAPF is the only evaluated approach that couples every detected hazard to a bounded, metric-neutral corrective action while maintaining near-goal stability, with zero clamp events in both scenarios, whereas CoT prompting increases from 9.7 to 14.0 events.
Abstract
Uncrewed aerial vehicles (UAVs) are increasingly deployed for autonomous navigation in complex outdoor environments, where dynamic conditions and mission requirements require intelligent adaptive decision-making. Existing optimization-based, Machine Learning (ML), and Reinforcement Learning (RL) approaches often rely on predefined models or task-specific training, limiting their generalization and adaptability in uncertain scenarios. Recent Large Language Model (LLM)-assisted approaches offer promising reasoning capabilities but remain constrained by limited agentic functionality, including insufficient memory, planning, and tool interaction mechanisms.This paper proposes an LLM-Agent-Based Path Finder (LAPF) framework for autonomous UAV navigation in town-scale outdoor environments. LAPF extends LLM-assisted navigation by integrating perception, memory, planning, and action modules into a closed-loop cognitive architecture. The proposed agent leverages prior navigation experiences, performs Chain-of-Thought (CoT) reasoning, couples each detected hazard to a bounded corrective action, and dynamically refines waypoint decisions based on environmental feedback.The three independent trials per method demonstrate that LAPF achieves mean path lengths of 512.83 m and 506.37 m, compared to the straight-line optimum of 497.33 m, corresponding to path length reductions of 17.2% and 15.6% relative to CoT prompting and absolute path efficiencies of 97.1% and 98.1% in open-field and obstacle-injected scenarios, respectively. Furthermore, LAPF is the only evaluated approach that couples every detected hazard to a bounded, metric-neutral corrective action while maintaining near-goal stability, with zero clamp events in both scenarios, whereas CoT prompting increases from 9.7 to 14.0 events.
Unmanned aerial vehicle vision-language navigation (UAV-VLN) requires agents to translate visual observations and language instructions into reliable flight actions in complex environments. Although recent end-to-end UAV vision-language-action (UAV-VLA) policies reduce reliance on separately designed perception, planning, and control modules, their behavior-cloning objectives provide limited corrective supervision for interactive closed-loop execution. Reinforcement learning (RL) offers a promising solution, while its effectiveness is constrained by inefficient use of samples, long-tailed scene distributions, and policy distribution shift during optimization. To this end, we propose RecoverFly, a failure-aware RL post-training framework for end-to-end UAV-VLA policies. Specifically, RecoverFly adapts token-level RL for stable optimization of grammar-constrained autoregressive UAV actions, revisits unresolved failure cases to strengthen corrective learning and sample utilization, and combines a two-stage long-tail scene curriculum with reference-policy regularization to improve scene adaptation while preserving acquired capabilities. Experiments on the TravelUAV benchmark demonstrate that RecoverFly achieves the best performance on the seen, unseen-map, and unseen-object splits. Moreover, compared to the AerialVLA initialization, RecoverFly improves success rate by 3.12 to 8.37 percentage points under a total rollout budget of about 30\% of the training-set size, validating its effectiveness, robustness, and generalization capabilities.
Boxiong Wang, Hui Kang, G. Sun et al.· 0 citations
Safe map-free multi-robot navigation requires robots to make real-time decisions from partial and noisy local observations in dynamic, unstructured environments. Existing approaches often depend on prior maps, computationally intensive perception pipelines, or carefully tuned interaction models, which limit their robustness on resource-constrained platforms. This paper proposes LSTP-Nav, a lightweight, decentralized navigation framework built on LSTP-Net that maps stacked 2D LiDAR observations, goal information, and velocity feedback directly to action. LSTP-Net combines GRU-based temporal modeling with attention to extract interaction-critical spatiotemporal cues at low computational cost. We further introduce an HS reward to provide smooth, heading-aware safety feedback, and develop PhysReplay-SimLab to improve training effectiveness through local replay of near-failure interactions. Extensive simulation and real-world experiments show that LSTP-Nav outperforms representative baselines, achieving 98.6%–100.0% success in single-robot settings and 97.8%–99.0% success in 10-robot settings with 5–35 obstacles. The policy also remains robust to LiDAR degradation, nonconvex layouts, random seeds, and large-scale density tests, while achieving over 40 Hz onboard inference on a Raspberry Pi 3 Model B with zero-shot sim-to-real transfer. Overall, LSTP-Nav provides a practical, deployable solution for real-time, map-free multi-agent navigation on computationally constrained mobile robots. Note to Practitioners—This study addresses a pressing practical need in large-scale site inspection and outdoor search-and-rescue: enabling mobile robots with limited onboard computation to navigate safely and efficiently in complex, previously unseen environments without relying on pre-built maps. In real deployments, robots must make decisions from partial and noisy sensor observations while interacting with moving objects (e.g., pedestrians or teammates), and conventional map-based pipelines or computationally heavy planners can become unreliable or infeasible. To tackle these constraints, we propose a lightweight spatiotemporal navigation policy that directly converts raw LiDAR observations into motion commands, combining temporal memory with attention to focus computation on interaction-critical cues. In addition, a heading-stability objective and a physics-based replay-assisted training workflow improve behavior smoothness and reduce failure cases such as oscillation, collisions, and getting trapped in cluttered areas. Extensive simulation and real-world tests show that the proposed approach improves navigation success and safety over representative baselines while maintaining real-time performance on CPU-only platforms, making it suitable for practical deployment on resource-constrained robots in dynamic, unstructured environments.
Xingrong Diao, Zhi-Qiang Sun, Jianwei Peng et al.· IEEE Transactions on Automat...· 0 citations
Cooperative unmanned aerial vehicle (UAV) path planning is one of the core challenges in the field of autonomous systems, requiring the simultaneous guarantee of safety, efficiency, and multi-agent collaborative consistency in continuously dynamic obstacle environments. Existing methods either rely on hand-crafted heuristic rules and thus lack adaptability, or employ deep reinforcement learning (DRL) but suffer from low sample efficiency and sparse reward problems, making it difficult to achieve high task success rates in complex dynamic scenarios. This paper proposes SkyAgent—a lightweight large language model (LLM)-driven hierarchical reinforcement learning framework specifically designed for adaptive cooperative path planning of two UAVs. SkyAgent adopts Qwen2.5-1.5B (Q4_K_M quantization, approximately 1 GB VRAM) as the high-level semantic planner, performing fully offline inference at approximately 1,900 ms model forward-pass per call on a local GPU via the Ollama framework without any external API; employs Proximal Policy Optimization (PPO) as the low-level continuous action controller; and achieves efficient decoupled coordination of the two-tier policy through an asynchronous sub-goal injection mechanism (replanning every K=20\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K=20$$\end{document} steps). In a continuous two-dimensional simulation environment of 100×100\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$100 \times 100$$\end{document} m2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^2$$\end{document} containing 8 static and 3 dynamic obstacles, SkyAgent is systematically evaluated against 7 baseline methods (Random, Greedy, APF, A*, PPO, SAC, OraclePID) over 30 random seeds. Experimental results demonstrate that SkyAgent achieves a mission success rate of 96.7%, comprehensively outperforming all baseline methods, improving by 16.7 percentage points over the vanilla PPO baseline, reducing collisions by 83%, shortening path length by 27 m, and improving sample efficiency approximately 3.3×\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times $$\end{document} (requiring only 55k training steps to reach the 80% success rate threshold). In the Dynamic-heavy scenario (30 obstacles), the success rate still reaches 90.9%, demonstrating significant generalization robustness. Six ablation experiments quantitatively validate the indispensability of the LLM planner, sub-goal guidance, sensor observations, cooperative reward, and reward shaping components. This work provides a feasible technical pathway and a reproducible evaluation benchmark for the collaborative deployment of lightweight LLMs and RL on GPU-equipped resource-constrained edge devices; we note that the present evaluation is confined to a two-dimensional, idealized simulation and that on-board deployment presupposes a GPU-class accelerator rather than an MCU-class micro-UAV.
Yuting Cao, Zheng Zhao, Jiekai Wu et al.· Journal of King Saud Univers...· 0 citations
The problems of partial observability and sensor shortage pose a significant challenge for autonomous Unmanned Aerial Vehicles (UAVs) as they prove to be challenging for conventional Deep Reinforcement Learning (DRL) methods to undertake well under such conditions. In this paper, a memory-augmented Proximal Policy Optimization (PPO) model extended using a Long Short-Term Memory (LSTM) network is proposed as a solution to such challenges. The observation space is constructed from 2D LiDAR and Inertial Measurement Unit (IMU) data to sense simultaneously external observation and internal state of motion, whereas the action space consists of continuous velocity commands. A shaped reward function is optimized for encouraging safe target approaching, obstacle avoidance, and convergence speed. Experimental outcomes show that the PPO-LSTM described herein achieves smoother paths, more robust reward convergence, and a much lower rate of collision than regular PPO. It also generalizes to new environments with movable obstacles. Qualitatively, the success rate increased from 64.5% to 83.9%, collision frequency reduced by over 70%, and path efficiency increased from 0.60 to 0.85, without suffering from unstable training behavior
M. Haddad, Dhayaa Khudher· Kufa journal of Engineering· 0 citations
: Autonomous navigation poses a key challenge in Artificial Intelligence (AI), necessitating agents to plan and execute actions in complex, partially visible surroundings. Simultaneous Localization and Mapping (SLAM) facilitates autonomous navigation of robots and vehicle objects to construct an unfamiliar environment map while concurrently monitoring their inside position. This systematic review investigates the nascent convergence of agentic AI, defined by goal-oriented autonomy, with adaptive decision-making and reasoning, with SLAM-based navigation systems. This paper utilized Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology, which concentrated on peer-reviewed articles published in (2017–2026), particularly in SLAM-based intelligent agents utilized for autonomous navigation. SLAM has transformed from a geometry-based localization framework into an advanced perceptual and reasoning paradigm for autonomous navigation. Recent advancements in agentic AI, semantic perception, multimodal learning, and embodied foundation models have facilitated autonomous agents in progressing from passive mapping to context-aware decision-making and goal-directed navigation. The emergence of agentic AI and embodied AI has revolutionized SLAM into a spatial world model that facilitates perception, memory, reasoning, and autonomous decision-making. Contemporary research emphasizes lifelong SLAM, collaborative multi-agent mapping, semantic world modelling, and the integration of Large Language Models (LLMs) and Vision-Language Models (VLMs) for intelligent autonomous agents. Consequently, SLAM has evolved from a localization instrument to an extensive cognitive framework facilitating advanced autonomous navigation systems.
Intelligent robot navigation in dynamic environments remains one of the most challenging problems in autonomous robotics because navigation systems must continuously perceive environmental changes, predict moving obstacles, and generate safe trajectories while maintaining operational efficiency. Traditional navigation approaches, including graph-based path planning, rule-based obstacle avoidance, and probabilistic localization, often exhibit limited adaptability when environmental conditions change rapidly. Recent advances in artificial intelligence, particularly Deep Reinforcement Learning (DRL), have enabled autonomous robots to learn navigation policies directly from environmental interactions without relying exclusively on handcrafted rules. DRL integrates perception, decision-making, and continuous learning into a unified framework, making it particularly suitable for complex and uncertain environments such as warehouses, hospitals, manufacturing plants, urban streets, and disaster-response scenarios.
This research-review paper presents a comprehensive analysis of Deep Reinforcement Learning-based intelligent robot navigation with emphasis on dynamic obstacle avoidance, adaptive path planning, perception integration, reward optimization, and policy learning. The paper synthesizes contemporary studies related to artificial intelligence, semantic decision intelligence, cyber-physical systems, cloud intelligence, autonomous optimization, and intelligent infrastructure to establish a multidisciplinary understanding of modern robotic navigation. Particular attention is devoted to semantic AI-enabled decision intelligence, which enhances contextual understanding during navigation and improves policy robustness in continuously evolving environments (Goyal, 2025).
A conceptual DRL navigation framework is proposed comprising environmental perception, state representation, policy optimization, experience replay, reward engineering, semantic reasoning, and adaptive trajectory generation. The framework demonstrates how semantic knowledge, sensor fusion, and reinforcement learning cooperate to produce robust navigation strategies under uncertainty. Furthermore, the paper evaluates challenges involving sparse rewards, safety constraints, computational complexity, sim-to-real transfer, multi-agent coordination, and real-time deployment.
Hiroshi Tanaka· European International Journ...· 0 citations