robot navigation

To address autonomous driving safety challenges in scenarios with high densities of vulnerable road users (VRUs) and edge cases such as construction zones and emergency response, this paper proposes a dual-mode vehicle system integrating autonomous navigation and remote takeover. Methodologically: (1) we design a topology-driven model predictive control (T-MPC) motion planner that generates parallel multi-strategy trajectories under joint probabilistic VRU collision constraints to enhance safety; (2) we introduce a novel vision–haptics fusion teleoperation interface to improve human–machine collaboration in edge scenarios; and (3) we integrate multi-sensor fusion perception, robust SLAM-based mapping, and closed-loop vehicle control. Experimental results demonstrate that the system achieves significantly higher safety and traffic efficiency than three baseline approaches in simulation; real-world closed-course tests successfully validate stable operation in both autonomous and remote-controlled modes.

Traditional navigation systems rely on LiDAR for geometric perception alone, lacking semantic understanding and thus failing to identify context-critical objects (e.g., scattered important documents), which compromises safety and practicality. To address this, we propose a lightweight framework that tightly integrates visual semantic perception with online A* path planning. Our method employs an embedded-friendly semantic segmentation model to detect user-defined visual constraints in real time; these are projected as dynamic non-geometric obstacles and fused with geometric sensor data to continuously update a global semantic map. This map drives online A* planning for context-aware, real-time obstacle avoidance. To the best of our knowledge, this is the first work to achieve closed-loop navigation on low-cost embedded platforms wherein obstacle definitions are dynamically reconstructed based on visual context. Experiments in simulation and on real robotic platforms demonstrate low latency, high robustness, and significantly improved safety and task adaptability for service robots operating in complex, unknown environments.

This work addresses the problem of local safe navigation for Ackermann-steered robots in mapless environments without global goals. The authors propose a real-time, perception-only obstacle avoidance method that identifies the largest open sector ahead to determine a safe heading and constructs left–right boundary constraints. A convex quadratic program is then employed to maximize clearance between the vehicle and surrounding obstacles, and a feedback linearization controller tracks the resulting smooth reference trajectory. The approach achieves high computational efficiency while ensuring both safety and trajectory smoothness. Experimental results demonstrate that, compared to existing exploration-based planners, the proposed method significantly reduces computation time and yields safer, more reliable paths.

Safe navigation for mobile robots in dynamic, uncertain environments remains challenging due to sensor noise and state estimation uncertainty, which traditional Control Barrier Functions (CBFs) fail to address probabilistically. Method: This paper proposes Distributionally Robust Control Barrier Functions (DR-CBFs), the first framework to directly incorporate raw sensor noise and state estimation uncertainty into probabilistic safety constraints—thereby relaxing the conventional reliance on deterministic models. DR-CBF integrates distributionally robust optimization with CBFs and Control Lyapunov Functions (CLFs), supporting robots with non-convex geometries and control-affine dynamics, while enabling real-time, closed-loop safety-critical control synthesis. Results: Evaluated in simulation and on a differential-drive robot platform, DR-CBF achieves strict probabilistic safety guarantees within millisecond-level control cycles, significantly improving navigation robustness and real-time performance in complex, dynamic scenarios.

To address safety and adaptability challenges in autonomous mobile robot navigation within unknown dynamic environments, this paper proposes a planning–control co-design framework. Methodologically, it integrates directional distance metrics with conical motion prediction to construct a risk assessment model, designs a customized cost map, and combines a reference corrector with control barrier functions (CBFs) to enable adaptive velocity modulation and trajectory tracking under safety-boundary constraints. Key contributions include: (i) the first coupling of directional distance and conical prediction for quantitative dynamic risk assessment; and (ii) a novel joint regulation mechanism integrating safety boundaries and reference correction. Extensive evaluations in simulation and complex real-world scenarios demonstrate that the proposed method significantly improves narrow-passage traversal success rates and reactive obstacle avoidance capability, achieving a favorable balance between high safety assurance and navigation efficiency.

Learning-based black-box autonomous mobile robots struggle to satisfy dynamically evolving human safety requirements. Method: This paper proposes a regulator-driven, post-hoc safety assessment framework. Its core innovations include: (i) systematically modeling human safety requirements as Signal Temporal Logic (STL) specifications for the first time; (ii) introducing differentiable, quantitative safety metrics—Total Robustness Value (TRV) and Local Robustness Value (LRV); and (iii) enabling closed-loop model retraining via external trajectory verification and robustness feedback. Results: In virtual driving tasks, speeding and lane-deviation violations decreased by 177% and 1138%, respectively. In robot navigation experiments, sharp-turn evasive capability improved by 300%, and time-to-collision with obstacles reduced by 49%. Real-world robotic deployment validates both effectiveness and generalizability.

This study addresses socially compliant navigation for mobile robots in human-dense environments. We systematically survey deep reinforcement learning (DRL)-based socially aware navigation methods and propose a unified analytical framework comparing value-based, policy-based, and actor-critic algorithms, advocating hybrid architectural designs. Our approach integrates diverse neural architectures—including feedforward networks, RNNs, CNNs, graph neural networks, and Transformers—to jointly model proxemics, human intent prediction, and quantitative comfort assessment. We innovatively establish a multidimensional benchmark balancing technical performance (e.g., collision rate, success rate) and human-centered metrics (e.g., perceived safety, social acceptability), empirically validating DRL’s efficacy in enhancing both navigation safety and human acceptance. Furthermore, we identify critical challenges—including sim-to-real transfer difficulty and the lack of standardized evaluation protocols—thereby providing theoretical foundations and practical guidelines for trustworthy social navigation research. (149 words)

To address the challenge of joint avoidance of dynamic and static obstacles in safety-critical navigation, this paper proposes a real-time motion planning framework based on LiDAR point clouds. Methodologically, it integrates dynamic obstacle tracking and mapping: point cloud segmentation and Kalman filtering jointly estimate and predict dynamic object states; these predictions are innovatively fused with static environment geometry to construct a forward-looking spatiotemporal occupancy map. Real-time collision checking over this map enables the synthesis of Control Barrier Function (CBF)-based safety constraints, which are embedded into a Nonlinear Model Predictive Control (NMPC) formulation for optimal trajectory generation. Experiments in both simulation and real-world deployments demonstrate significant improvements in obstacle avoidance safety and robustness over state-of-the-art baselines. The implementation is open-sourced.

Social robots in crowd navigation face a bidirectional intent opacity problem: robots struggle to infer human willingness to cooperate, while humans cannot interpret robot motion planning—leading to navigation instability under unexpected interactions. To address this, we propose a communication-triggering mechanism grounded in geometric context and cooperative intent assessment. It establishes a quantifiable evaluation framework distinguishing cooperative from non-cooperative pedestrians, integrating head orientation geometry, behavior prediction models, and socially aware navigation to dynamically determine interaction timing and generate appropriate verbal or gestural responses. Our key contribution lies in modeling cooperative intent as a computable geometric-behavioral coupling feature—eschewing hand-crafted rules. Experiments demonstrate significant improvements in navigation fluency and naturalness, with a 23.6% increase in task success rate in high-density, dynamic environments.

This work addresses the challenge of safe and efficient navigation for micro air vehicles (MAVs) in dynamic human crowds—specifically requiring full-body human motion modeling beyond root-joint trajectory tracking. We propose a model predictive control (MPC) framework that integrates formal guarantees with data-driven learning. Our key contributions are: (i) a reachability-based safety constraint mechanism that imposes explicit bounds only on the initial control input, yet rigorously captures its propagated effect over the entire prediction horizon—ensuring strict safety while significantly reducing conservatism; and (ii) tight coupling with a data-driven human motion predictor, trained on real-world pedestrian trajectories, enabling multi-task capabilities including goal-directed navigation and human tracking. Extensive simulations and real-world flight experiments demonstrate superior performance over state-of-the-art baselines in safety compliance, navigation efficiency, and task adaptability.