Cognitive Robotics最新文献

Knee joint rehabilitation exercise refers to a therapeutic procedure of a patient having dysfunctions in certain abilities to move knee joint due to some medical conditions like trauma or paralysis. The exercise is basically a series of repeated assistive physical movements within the range of motion (RoM) of the joint. Reflex action of limbs during RoM exercise causes inappropriate balance of load which may cause secondary injuries, such as damages of muscle or tendon tissues. Establishing correlation between impedance data and limb motions is important to solve this problem. This paper aims to design and modeling of a robotic arm with an original approach in control strategy which is developed based on the correlation in between the joint-impedances and joint-motion characteristics during exercise. The knee joint impedances are estimated based on the internal feedback of the system dynamics, that lead to design the torque compensator to improve the overall control signals in real time. This paper also demonstrates the characteristics of various responses of the system during exercise with human subject. Results have reflected good performances with low position and velocity tracking errors, $\pm 0.{02}^{\circ }$ and $0.04rad.se{c}^{-1}$ during hold phase; and $\pm 0.{14}^{\circ }$ and $0.17rad.se{c}^{-1}$ during motion phse. Though, the limitation of the prototype is its current RoM (limited to $0^{\circ }$–${25}^{\circ }$), the system has potential in the application of RoM exercise for paraplegic or monoplegic patients.

Mobile robot path planning involves decision-making in uncertain, dynamic conditions, where Reinforcement Learning (RL) algorithms excel in generating safe and optimal paths. The Deep Deterministic Policy Gradient (DDPG) is an RL technique focused on mobile robot navigation. RL algorithms must balance exploitation and exploration to enable effective learning. The balance between these actions directly impacts learning efficiency.

This research proposes a method combining the DDPG strategy for exploitation with the Differential Gaming (DG) strategy for exploration. The DG algorithm ensures the mobile robot always reaches its target without collisions, thereby adding positive learning episodes to the memory buffer. An epsilon-greedy strategy determines whether to explore or exploit. When exploration is chosen, the DG algorithm is employed. The combination of DG strategy with DDPG facilitates faster learning by increasing the number of successful episodes and reducing the number of failure episodes in the experience buffer. The DDPG algorithm supports continuous state and action spaces, resulting in smoother, non-jerky movements and improved control over the turns when navigating obstacles. Reward shaping considers finer details, ensuring even small advantages in each iteration contribute to learning.

Through diverse test scenarios, it is demonstrated that DG exploration, compared to random exploration, results in an average increase of 389% in successful target reaches and a 39% decrease in collisions. Additionally, DG exploration shows a 69% improvement in the number of episodes where convergence is achieved within a maximum of 2000 steps.

Stroke is a leading cause of neurological disorders that result in physical disability, particularly among the elderly. Neurorehabilitation plays a crucial role in helping stroke patients recover from physical impairments and regain mobility. Physical therapy is one of the most effective forms of neurorehabilitation, but the growing number of patients requires a large workforce of trained therapists, which is currently insufficient. Robotic rehabilitation offers a promising alternative, capable of supplementing or even replacing human-assisted physical therapy through the use of rehabilitation robots. To design effective robotic devices for rehabilitation, a solid foundation of knowledge is essential. This article provides a comprehensive overview of the key elements needed to develop human upper extremity rehabilitation robots. It covers critical aspects such as upper extremity anatomy, joint range of motion, anthropometric parameters, disability assessment techniques, and robot-assisted training methods. Additionally, it reviews recent advancements in rehabilitation robots, including exoskeletons, end-effector-based robots, and planar robots. The article also evaluates existing upper extremity rehabilitation robots based on their mechanical design and functionality, identifies their limitations, and suggests future research directions for further improvement.

Path Planning in a collaborative mobile robot system has been a research topic for many years. Uncertainty in robot states, actions, and environmental conditions makes finding the optimum path for navigation highly challenging for the robot. To achieve robust behavior for mobile robots in the presence of static and dynamic obstacles, it is pertinent that the robot employs a path-finding mechanism that is based on the probabilistic perception of the uncertainty in various parameters governing its movement. Partially Observable Markov Decision Process (POMDP) is being used by many researchers as a proven methodology for handling uncertainty. The POMDP framework requires manually setting up the state transition matrix, the observation matrix, and the reward values. This paper describes an approach for creating the POMDP model and demonstrates its working by simulating it on two mobile robots destined on a collision course. Selective test cases are run on the two robots with three categories – MDP (POMDP with belief state spread of 1), POMDP with distribution spread of belief state over ten observations, and distribution spread across two observations. Uncertainty in the sensor data is simulated with varying levels of up to 10 %. The results are compared and analyzed. It is demonstrated that when the observation probability spread is increased from 2 to 10, collision reduces from 34 % to 22 %, indicating that the system's robustness increases by 12 % with only a marginal increase of 3.4 % in the computational complexity.

In recent years, the rapid industrialization of the world has led to an increasing importance of energy minerals. However, due to the scarcity of mineral resources, opportunities to rely on alternative energy are escalating. As a result, exploration of ocean resources, which exist abundantly in the sea, is being pursued. However, the manual exploration of ocean resources by diving and visually searching is dangerous and impractical. Therefore, it is pertinent to safely advance underwater exploration by having robots perform the work instead. In underwater environments, robots are commonly used as a mainstream exploration tool due to the various hazardous environmental conditions. However, there are several problems with controlling robots in underwater environments, and one of them is poor visibility underwater. Therefore, to improve visibility underwater, efforts are being made to achieve high resolution using super-resolution technology on underwater images. In this paper we first introduce the general model and architecture in GAN. Then we combine the GAN modal and characteristics of the underwater environment, elaborating how ESRGAN can be suitable for such circumstance. For data from ECCV2018 PIRM-SR, ESRGAN outperforms other traditional model like EnhanceNet [1], EDSR [2], RCAN [3], at least 24 % [4]. Such model can be equipped with robotics that highly depends on the resolution of the image, such as autonomous underwater vehicles (AUVs) and remotely operated vehicles (ROVs).

UAVs are often limited by limited resources when performing flight tasks, especially the contradiction between storage resources and computing resources when the huge YOLOv3 model is deployed on the edge UAVs. In this paper, we tend to compress YOLOv3 model in different aspects to achieve load availability at the edge. In this paper, deep separable convolution is introduced to reduce the computation of the model. Then, PR regularization term is used as the regularization term of sparse training to better distinguish scaling factors, and then the hybrid pruning combining channel pruning and layer pruning is carried out on the model according to scaling factors, in order to reduce the number of model parameters and the amount of calculation. Finally, since the training data is a 32-bit floating point number, DoReFa-Net quantization method is used to quantify the model, so as to compress the storage capacity of the model. The experimental results show that the compression scheme proposed in this paper can effectively reduce the number of parameters by 97.5 % and the calculation amount by 82.3 %, and can maintain the original detection efficiency of UAVs.