Aug 2024· IEEE Transactions on Automation Science and Engineering· Vol 23, pp. 14837-14851· 0 citations· 52 references
Computer ScienceEngineering
TL;DR
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 and introduces an HS reward to provide smooth, heading-aware safety feedback, and develops PhysReplay-SimLab to improve training effectiveness through local replay of near-failure interactions.
Abstract
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.
Autonomous navigation in real-world public service, industrial inspection, and emergency response often faces frequent changes in nominally static scene structures, which can quickly invalidate pre-built global maps and naturally lead to a mapless navigation setting. We propose an end-to-end 3D LiDAR based navigation framework that directly maps raw point clouds and relative goal cues to discrete actions. First, an importance gaze point cloud representation (IGPR) converts unordered sparse scans into a compact two-channel image representation via view-region enhancement and an adaptive reciprocal factor, improving sensitivity to decision critical geometry. Second, a geometry-driven dense reward is designed from raw point cloud structure and goal relative states to accelerate reinforcement learning and mitigate local-optimum behaviors. Third, a raw point- cloud driven Gaussian Control Barrier Functions (RPGC) safety controller performs minimal intervention when unsafe proximity is detected, improving deployment robustness without overriding the learned policy. Extensive simulations across diverse obstacle styles and time varying layouts demonstrate consistently high success rates and stable path efficiency. Real world experiments with novel obstacle geometries not encountered in simulation further validate direct sim-to-real transfer of the learned policy without additional real-world training.
Yue Zhai, Yanzi Miao· IEEE Robotics and Automation...· 0 citations
Mapless navigation often removes global maps while retaining localization-derived goal vectors or bearings. We study a stricter setting in which a mobile robot observes only local LiDAR, scalar goal range, and short histories of executed actions; neither pose nor goal direction is provided to the policy. We introduce ACR-Nav, an action-conditioned range navigation framework that converts scalar-range evolution into closed-loop progress information. Its range–action history associates each distance change with the motion that produced it, while the sectorized LiDAR captures local geometry and short-term obstacle motion. A LiDAR-only safety filter provides immediate collision intervention, and a static-to-mixed curriculum stabilizes learning. A lightweight multilayer–perceptron is optimized with Proximal Policy Optimization (PPO), while the ACR-Nav formulation itself remains optimizer-agnostic. In corridor simulations, ACR-Nav achieved 93.2%, 80.4%, and 84.4% success in static, mixed, and dynamic environments. Removing the safety filter reduced success by 15.2, 14.6, and 16.0 percentage points in static, mixed, and dynamic environments, respectively, and random-goal tests yielded 91.2% and 81.4% success in static and mixed settings. Topology-shift experiments further quantified adaptation to an L-shaped corridor. The results show that action-conditioned scalar-range evolution can support goal-directed, segment-level navigation within locally straight corridor passages without exposing robot pose or target bearing to the policy.
This paper presents an observation-only autonomy framework for Unmanned Underwater Vehicles (UUVs) navigation in dynamic underwater environments that integrates persistent occupancy mapping, global clearance-aware planning, and risk-aware local control. The proposed pipeline constructs occupancy maps solely from onboard sonar and depth image observations, adapts a clearance-constrained global planner (GP) to provide long-horizon structure, and integrates a reinforcement learning (RL) policy to handle short-range tracking and reactive avoidance. To further support decision-making under partial observability, the system learns a compact latent state representation from onboard sensor data, encoding environmental structure, obstacle dynamics, and uncertainty. Behavior tree (BT) distillation with staged supervision is introduced to improve safety and training stability, while an uncertainty-calibrated distillation mechanism reweights teacher guidance using online latent-model uncertainty, emphasizing uncertain regimes during learning, with time-to-collision (TTC) and clearance cues remaining explicit in planning and local policy features. To demonstrate the efficacy of the framework, a reproducible multi-seed evaluation protocol is established in high-fidelity GPU-accelerated simulation using NVIDIA Isaac Sim, and performance is benchmarked against BT-only and standard RL baselines. The results obtained demonstrate improved robustness and safety under dynamic conditions, thus providing a general pipeline with a unified hybrid planning learning architecture and a reproducible methodology for robust UUV autonomy under partial observability.
This work provides a feasible technical pathway and reproducible evaluation benchmark for the collaborative deployment of lightweight LLM planner, sub-goal guidance, sensor observations, cooperative reward, and reward shaping components and quantifies the indispensability of the LLM planner.
Yuting Cao, Zheng Zhao, Jiekai Wu et al.· Journal of King Saud Univers...· 0 citations
Energy-efficient local navigation on skid-steer mobile robots (SSMRs) in dense environments is constrained by a structural trade-off: the low-curvature, low-power manoeuvres that minimise locomotion energy conflict with the high-curvature, high-yaw-rate corrections required to weave through dense obstacle fields without colliding. Model-based planners do not resolve this tension but expose it as a cost knob, since weighting energy more heavily cuts consumption only by degrading performance as the scenes grow harder. We present LEAN, a deep reinforcement learning local planner that learns locomotion-efficient navigation from a calibrated energy signal while being model-free at inference, mapping observations directly to commands without solving any closed-form dynamic or power model online as model-based planners do. The energy signal enters the policy only as raw motor currents, and in place of dense goal-progress shaping the policy is guided by hindsight relabeling that keeps the reward path-shape-agnostic. Benchmarked on a 140-kg tracked SSMR against NMPC and EA-DWA variants that share the simulator’s identified dynamics and power model, raising the energy weight of either family lowers consumption only by sacrificing success, leaving the most aggressive energy-aware baselines up to 14 success points below LEAN at the hard tier. LEAN instead reduces energy by 10 to 14% in simulation and 19% on hardware over the energy-blind LEAN-Ablation at unchanged success rate and selects actions more than an order of magnitude faster than the model-based controllers, with consistent sim-to-real transfer in cluttered outdoor environments. The planner is named EA-SAC in the title and LEAN in the text, figures and tables; both names denote the same method.
Robot crowd navigation requires safe and efficient decision-making under dense, dynamic, and multimodal human--robot interactions. Existing reinforcement-learning methods typically output a single reactive action at each timestep, which limits their ability to represent diverse short-term avoidance strategies. We propose Planning Diffusion Policy Optimization (PDPO), an offline-to-online reinforcement-learning framework that uses a diffusion policy to generate short-horizon action chunks for crowd navigation. PDPO is first pretrained on collision-avoidance demonstrations and then fine-tuned online with PPO by treating the denoising process as an internal decision process. During execution, the policy generates a five-step action chunk and applies it in a receding-horizon manner. Furthermore, we observe an evaluation artifact in common crowd-navigation benchmarks: without explicit boundary constraints, learned agents may leave the valid domain and bypass dense crowds. To address this, we introduce a setting in which boundary violations are treated as collisions. Experiments show that PDPO obtains an improved success rate over strong baselines, and ablations demonstrate that action chunks are especially important for the modified bounded benchmark.
Wen-Dong Li, J. Garcke· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
