Motion-planning algorithms play a vital role in attaining various levels of autonomy for any ground or flying agent. Three-dimensional (3D) motion-planning is interesting, but rather complex, especially for flying agents such as autonomous unmanned aerial vehicles (UAVs), due to the increased dimensionality of space and consideration of dynamical constraints for a feasible trajectory. Usually, in 3D path search problems, it is hard to avoid extra expanded nodes due to increased dimensionality with various available search options. Therefore, this paper discusses and implements a modified heuristic-based A* formalism that uses a truncation mechanism in order to eradicate the mentioned problem. In this formalism, the complete motion planning is divided into shortest path search problem and smooth trajectory generation. The shortest path search problem is subdivided into an initial naïve guess of the path and the truncation of the extra nodes. To find a naïve shortest path, a conventional two-dimensional (2D) A* algorithm is augmented for 3D space with six-sided search. This approach led to the inclusion of extra expanded nodes and, therefore, it is not the shortest one. Hence, a truncation algorithm is developed to further process this path in order to truncate the extra expanded nodes and then to shorten the path length. This new approach significantly reduces the path length and renders only those nodes that are obstacle-free. The latter is ensured using a collision detection algorithm during the truncation process. Subsequently, the nodes of this shortened path are used to generate a dynamically feasible and optimal trajectory for the quadrotor UAV. Optimal trajectory generation requires that the plant dynamics must be differentially flat. Therefore, the corresponding proof is presented to ensure generation of the optimal trajectory. This optimal trajectory minimizes the control effort and ensures longer endurance. Moreover, quadrotor model and controllers are derived as preliminaries, which are subsequently used to track the desired trajectory generated by the trajectory planner. Finally, results of numerical simulations which ultimately validate the theoretical developments are presented.

Autonomous navigation in unknown environments is still a challenge for robotics. Many efforts have been exerted to develop truly autonomous goal-oriented robot navigation models based on the neural mechanism of spatial cognition and mapping in animals’ brains. Inspired by the Semantic Pointer Architecture Unified Network (SPAUN) neural model and neural navigation mechanism, we developed a brain-like biologically plausible mathematical model and applied it to robotic spatial navigation tasks. The proposed cognitive navigation framework adopts a one-dimensional ring attractor to model the head-direction cells, uses the sinusoidal interference model to obtain the grid-like activity pattern, and gets optimal movement direction based on the entire set of activities. The application of adaptive resonance theory (ART) could effectively reduce resource consumption and solve the problem of stability and plasticity in the dynamic adjustment network. This brain-like system model broadens the perspective to develop more powerful autonomous robotic navigation systems. The proposed model was tested under different conditions and exhibited superior navigation performance, proving its effectiveness and reliability.

Robotic grasping in cluttered environments is a fundamental and challenging task in robotics research. The ability to autonomously grasp objects in cluttered scenes is crucial for robots to perform complex tasks in real-world scenarios. Conventional grasping is based on the known object model in a structured environment, but the adaptability of unknown objects and complicated situations is constrained. In this paper, we present a robotic grasp architecture of attention-based deep reinforcement learning. To prevent the loss of local information, the prominent characteristics of input images are automatically extracted using a full convolutional network. In contrast to previous model-based and data-driven methods, the reward is remodeled in an effort to address the sparse rewards. The experimental results show that our method can double the learning speed in grasping a series of randomly placed objects. In real-word experiments, the grasping success rate of the robot platform reaches 90.4%, which outperforms several baselines.

In unstructured environments, robots need to deal with a wide variety of objects with diverse shapes, and often, the instances of these objects are unknown. Traditional methods rely on training with large-scale labeled data, but in environments with continuous and high-dimensional state spaces, the data become sparse, leading to weak generalization ability of the trained models when transferred to real-world applications. To address this challenge, we present an innovative maximum entropy Deep Q-Network (ME-DQN), which leverages an attention mechanism. The framework solves complex and sparse reward tasks through probabilistic reasoning while eliminating the trouble of adjusting hyper-parameters. This approach aims to merge the robust feature extraction capabilities of Fully Convolutional Networks (FCNs) with the efficient feature selection of the attention mechanism across diverse task scenarios. By integrating an advantage function with the reasoning and decision-making of deep reinforcement learning, ME-DQN propels the frontier of robotic grasping and expands the boundaries of intelligent perception and grasping decision-making in unstructured environments. Our simulations demonstrate a remarkable grasping success rate of 91.6%, while maintaining excellent generalization performance in the real world.

In an unstructured environment, multiple objects are closely arranged, which leads to the complexity of robot grasping. Deep reinforcement learning(DRL) can take long-term reasoning, and make better and more robust decisions in a series of decisions, allowing the robot to continuously learn from its own experience. In this paper, we demonstrate that a new deep reinforcement learning algorithm based on Q-learning and attention mechanisms can effectively solve the robot grasping problem. In addition, we present a grasp and move strategy that reduces complexity by replacing hand-designed features with data-driven ones. The robot is used to move the objects to create a larger grasping space when the objects are densely distributed to improve the grasping success rate. The application of the proposed model on a simulated UR5 robot manipulator demonstrates the effectiveness and stability of the model. The results show that our proposed method outperforms the baseline framework in terms of grasping accuracy and efficiency.