Current robotic studies are focused on the performance of specific tasks. However, such tasks cannot be generalized, and some special tasks, such as compliant and precise manipulation, fast and flexible response, and deep collaboration between humans and robots, cannot be realized. Brain-inspired intelligent robots imitate humans and animals, from inner mechanisms to external structures, through an integration of visual cognition, decision making, motion control, and musculoskeletal systems. This kind of robot is more likely to realize the functions that current robots cannot realize and become human friends. With the focus on the development of brain-inspired intelligent robots, this article reviews cutting-edge research in the areas of brain-inspired visual cognition, decision making, musculoskeletal robots, motion control, and their integration. It aims to provide greater insight into brain-inspired intelligent robots and attracts more attention to this field from the global research community.

Since space intelligent robots are not restricted by physiological conditions, it is an attractive choice for the development of automation technology to use them for space exploration and utilization. It is currently the key development direction of the major space powers over the world. This paper first investigates the robotic manipulators and humanoid robot systems for space station applications and reviews theories and methods for robots to achieve large-range stable motion and intelligent dexterous manipulation. Then, the intelligent robot systems for on-orbit satellite maintenance are reviewed, and the related technologies of multirobot collaboration are analyzed. Finally, we investigate the intelligent robot systems for on-orbit assembly of large-scale spatial structures and summarize the technologies of modular assembly and on-orbit manufacture. Overall, this paper reviews the technological progress and development trends of space robots, which provides a good reference for further technical research in this field.

Based on basic emotion modulation theory and the neural mechanisms of generating complex motor patterns, we introduce a novel emotion-modulated learning rule to train a recurrent neural network, which enables a complex musculoskeletal arm and a robotic arm to perform goal-directed tasks with high accuracy and learning efficiency. Specifically, inspired by the fact that emotions can modulate the process of learning and decision making through neuromodulatory system, we present a model of emotion generation and modulation to adjust the parameters of learning adaptively, including the reward prediction error, the speed of learning and the randomness in action selection. Additionally, we use Oja learning rule to adjust the recurrent weights in delayed-reinforcement tasks, which outperforms the Hebbian update rule in terms of stability and accuracy. In the experimental section, we use a musculoskeletal model of the human upper limb and a robotic arm to perform goal-directed tasks through trial-and-reward learning respectively. The results show that emotion-based methods are able to control the arm with higher accuracy and a faster learning rate. Meanwhile, emotional Oja agent is superior to emotional Hebbian one in term of performance.

Safe and efficient coordinated motion planning is crucial for the dual-arm robot's manipulation in the constrained environment. This article studies a fast optimization-based coordinated motion planning method for dual-arm robots based on a parallel differential dynamic programming (DDP) solver. First, the proposed method models the unstructured environment using swept sphere volumes and turns the safety and task constraints, including collision avoidance, path constraints, and closed kinematics constraints, into mathematical equalities and inequalities in an optimization problem. Second, to efficiently solve this high-dimensional and multiconstraint optimization problem, a parallel constrained differential dynamic programming (PC-DDP) solver is developed based on a new multiple shooting strategy. It utilizes the augmented Lagrangian method to handle various constraints and achieves high solving efficiency using the parallelization strategy with an approximation and the serial correction of the value function. Simulations and experiments of opening doors and transferring the box are conducted on the robot with dual 7-DOF manipulators. The results show that the proposed method works well in the motion planning of coordinated manipulations in constrained environments and that the PC-DDP is much faster than other DDP variants and large-scale optimization solvers.

Musculoskeletal robots with human-like joints, muscles, and actuation mechanisms are characterized by exceptional dexterity, compliance, and versatility. However, existing reinforcement learning methods for such robots rely on precise and sufficient state observation, rendering them vulnerable to perturbations. To address this limitation, this paper proposes a robust motion learning method based on a recurrent neural network (RNN) and muscle synergy. First, the proposed method utilizes task-joint-muscle space states to create an RNN-based neuromuscular controller. Furthermore, a motion learning method with a synergistic constraint of muscles is developed. Additionally, theoretical analysis confirms that the RNN-based controller is more robust to perturbations of state observation than a Multilayer Perceptron (MLP) based controller. The proposed method is evaluated on a simulated musculoskeletal robot and demonstrates superior robustness to other MLP-based reinforcement learning methods. Furthermore, the proposed method is also validated on a musculoskeletal robot hardware system, indicating its potential for real-world applications. Note to Practitioners —Musculoskeletal robots have shown promising potential in various applications, but their development and application have been restricted by the limited robustness of existing reinforcement learning methods. These methods work well in simulation with precise and sufficient state observation, but exhibit a considerable degradation of performance in real-world environments with perturbed and insufficient state observation. In this article, we propose a novel method to improve the robustness of motion learning for musculoskeletal robots using a recurrent neural network and muscle synergy. The proposed method is theoretically and experimentally validated to perform well not only under precise and sufficient state observation but also in the presence of perturbed and insufficient state observation. Our results demonstrate the effectiveness of the proposed method and motivate the application of reinforcement learning methods to musculoskeletal robots. This research contributes to the advancement of robust motion learning for musculoskeletal robots and paves the way for wider adoption in real-world applications.

Immediate rewards are usually very sparse in the real world, which brings a great challenge to plain learning methods. Inspired by the fact that emotional reactions are incorporated into the computation of subjective value during decision-making in humans, an emotion-motivated decision-making framework is proposed in this paper. Specifically, we firstly build a brain-inspired computational model of amygdala-hippocampus interaction to generate emotional reactions. The intrinsic emotion derives from the external reward and episodic memory, and represents three psychological states: valence, novelty and motivational relevance. Then, a model-based decision-making approach with emotional intrinsic rewards is proposed to solve the continuous control problem of mobile robots. This method can execute online model-based control with the constraint of the model-free policy and global value function, which is conducive to getting a better solution with a faster policy search. The simulation results demonstrate that the proposed approach has higher learning efficiency and maintains a higher level of exploration, especially in some very sparse-reward environments.