Rui Lin

Instance Image Navigation (IIN), an embodied navigation task, requires an agent to autonomously explore unknown environments and navigate to the vicinity of a target object indicated by an image. The core challenges are efficient semantic exploration without prior maps and accurate target discrimination among similar objects. Existing methods are mainly divided into modular approaches and 3D Gaussian splatting methods; the former lack semantic guidance, while the latter rely on pre-built maps, unsuitable for online navigation. To address this, this paper proposes Flat-IIN, a modular framework using 2D multi-channel semantic maps and a vision-language model (VLM). It constructs multiple 2D maps and integrates Multi-View Active Semantic Exploration, Global Backtracking, and VLM-based Frontier Exploration. These collaboratively leverage the maps to accomplish IIN. Experimental results on the Habitat-Matterport 3D semantic dataset show that Flat-IIN achieves a success rate of 0.794 and success weighted by normalized inverse path length of 0.331. These results demonstrate competitive navigation performance and validate its effectiveness.

In this paper, we propose a novel dynamic path planning method for multiple mobile robots, named CBS-TLTD3. The method combines the Conflict-Based Search (CBS) algorithm for global path planning and enhances the obstacle avoidance capability of the Twin Delayed Deep Deterministic Policy Gradient (TD3) model through transfer learning, thereby developing the TLTD3 model. Additionally, we introduce an intermediate planner to coordinate these two methods, enabling continuous path planning. This integrated approach enables real-time conflict detection and resolution, ensuring stable and continuous global path planning. Furthermore, the effectiveness of this integrated method is validated through simulations and real-world experiments, providing valuable insights and solutions for multi-robot path planning applications.

In this article, we propose an autonomous exploration system based on transfer learning for target point exploration in unknown environments. The system generates a series of clustering points based on the local perceptual information and selects suitable local exploration points through a heuristic method to guide the robot towards the global target direction. This approach alleviates the problem of local optima to a significant extent. To reduce the time cost of deep reinforcement learning in the initial stages, we employ transfer learning by training a model capable of avoiding static obstacles in a simulated environment to accomplish local dynamic navigation tasks. By combining the locally learned dynamic navigation policy with global motion planning, we achieve autonomous exploration for the robot. During the fully autonomous navigation process, we record the robot’s movement trajectory and the generated map. Experimental results demonstrate that compared to similar exploration methods, this approach exhibits advantages in complex dynamic environments, even with the sole use of two-dimensional laser, without the need for maps or excessive information.

Robotic manipulators usually exhibit time‐varying, nonlinear, and coupled dynamics due to the parameter perturbations, disturbances, and other uncertainties. Traditional control algorithms typically do not possess parameters' self‐adaptive learning ability, limiting the tracking performance of the robot. In order to address these issues, an adaptive sliding mode control method based on reinforcement learning (ASMRL) is proposed in this paper, where a proportional–integral sliding mode (PISM) controller is used to address the nonlinear problem in the system, and deep deterministic policy gradient (DDPG) agent is adopted to conduct the parameters' learning of the PISM controller using its adaptive characteristics and the autonomous learning ability. The simulation results illustrate that the proposed method can effectively achieve better tracking performance compared with two other control methods, demonstrating the effectiveness of the proposed approach.

Multirobot path planning in complex environments is a challenging research area. This paper proposes a path planning method for multirobot systems based on distributed multiagent deep reinforcement learning. We propose a multirobot dynamic window approach (MRDWA) in which a central controller facilitates sensor information sharing among robots, enabling locally optimal path planning considering the behavior of other robots. We incorporate the output velocity information into the observation function to form an efficient and low-dimensional state representation. Additionally, we employ the multiagent deep deterministic policy gradient (MADDPG) reinforcement learning algorithm to directly map part of the observation information to motion commands for multiple robots, enabling effective obstacle avoidance strategies. An improved action module is developed by using velocity and angular velocity increments and an action selector to refine the output. Furthermore, we introduce a multirobot reward module utilizing heuristic functions to guide the robots to quickly and efficiently identify feasible paths. We also propose a multirobot dynamic constraint reward function to optimize the multirobot trajectories. The MRDWA-MADDPG algorithm is validated through simulations and real-world experiments, demonstrating its effectiveness in diverse complex multirobot path planning scenarios. Our method outperforms conventional algorithms in terms of success rate, arrival time, and overall decision making in complex scenarios. Moreover, our method has a faster computation speed and a shorter training time, produces smoother trajectories, and is easier to deploy on real robots than other learning-based methods.

In the article, a global localization algorithm based on improved ultra-wide-band-based adaptive Monte Carlo localization is proposed for quick and robust kidnap recovery of mobile robot. First, two ultra-wide-band modules, the tag installed inside the mobile robot and the anchor installed inside charging station, are used to obtain the relative distance between the mobile robot and the charging station. Second, the global grid map is converted into a map with obstacle noise given the ranging accuracy of the ultra-wide-band modules with different obstacles. Third, while the robot is kidnapped, matching grids are screened based on the range information of ultra-wide-band modules and the obstacle noise of the grids. Finally, global localization algorithm is performed based on ultra-wide-band-based adaptive Monte Carlo localization to convert randomly generated particles from the whole map into randomly generated particles in the local map. Experimental results based on gazebo simulation and a real scenario showed that our global localization algorithm based on improved ultra-wide-band-based adaptive Monte Carlo localization not only significantly helped to improve the chances of the robot global pose recovery from lost or kidnapped state but also enabled the robot kidnap recovery with a smaller number of randomly generated particles, thus reducing the time to recover its accurate global localization. The algorithm was also more effective especially for kidnap recovery in a similar and large scenario.

In some large environments where humans and machines coexist, such as smart factories, restaurants, and hotels, multiple mobile robots system have certain advantages over single mobile robot in terms of task complexity, execution efficiency, and system robustness. However, path conflicts and blockages may occur among robots without effective global coordination method, especially in some special areas like narrow straight roads (N-S-R) and crossroads (C-R). To solve this problem, using the global topology map might be a good solution. Therefore, this paper proposes a method called Surplus Topology Generation Algorithm (STGA) for constructing a global topological map based on grid map representation. On the basis of grid maps, we designed Node-Filter to generate enough topological nodes, and simplified the node set by Node-Selection. In special areas such as N-S-R and C-R, which are prone to path conflicts and blocking scenarios, we extracted the Harris corner point and combined them to select the corner point combinations that characterize some special features like passage entrance, etc. Then, feature nodes that can characterize the special features were obtained by establishing mapping relationships between nodes and corner point combinations. Finally, the global topology map was obtained after establishing the reachability relationships between nodes. The simulation experiment results showed that the topological maps generated by STGA had obvious advantages in the richness of topological relationships. Besides, the topological maps also showed certain advantages compared to grid maps for the planning efficiency and security of global paths. Furthermore, the rich topological nodes on the global topological map were feasible in solving path conflicts of multiple robots. Note to Practitioners—Multi-robot system has better efficiency and system robustness, but how to solve the path conflicts among robots is a problem. We propose a method to solve multi-robot conflicts with global topological maps. This topological map is constructed by STGA. The grid map environment is discretized into a set of topological nodes, which have been used to maintain the spatial transitions of each robot in the environment. When robots encounter unavoidable path conflicts, they are scheduled with the help of topological nodes to achieve orderly and efficient collaboration.