Search

Transcranial magnetic stimulation efficacy is largely dependent upon coil position and orientation. A good method for describing coil placement is required for both computational optimization (planning) and actual placement (implementation). In coordinate dependent parameter-spaces (CDPs), three-dimensional coordinates are used to represent coil position and three orthogonal unit vectors are used to represent coil orientation. A CDP can precisely describe arbitrary coil placement; therefore it offers great advantage in computational optimization which checks through all possible placements. However, a neuronavigation system is usually required to accurately implement the optimized CDP parameters on a participant’s head. Routine clinical practice, on the other hand, often uses the International 10-20 system to describe coil placement. Although the 10-20 system can only perform modeling and placement at limited scalp landmarks, it allows the synthesis of different individuals’ targeting effects to find group-optimal parameters; it also allows manual placement, which is important for commonly-seen use cases without individual MRI scans and navigation devices. This study proposes a new scalp geometry based parameter-space (SGP), integrating the advantages of CDP and 10-20 methods. Our SGP 1) can quantitatively specify all possible conventional coil positions and orientations on an individual’s scalp, which is important for electrical modeling and optimization, 2) maintains inter-individual correspondence, which is important for synthesizing TMS effects from different individuals and studies. 3) enables fast and simple manual implementation. Demonstration experiments were conducted to illustrate the application of an SGP-based framework for both individual and group-based optimization. A measurement experiment was performed to evaluate speed, precision and reliability of SGP-based manual implementation; results show it surpasses previous manual placement methods.

Programming control systems for mobile robots is complicated and time-consuming, due to three aspects, i.e., the robot behavior coordination, the distributed multi-robot cooperation and the robot software reusability. Subsumption model is a robust control architecture for mobile robots. ALLIANCE model extends it to multirobot systems, which is a fully distributed, fault-tolerant model. Robot operating system (ROS) provides a lot of reusable robot modules. By combining the above three, we propose a software framework named ALLIANCE-ROS for developing fault-tolerant cooperative multi-robot systems with abundant software resources available. We encapsulate the ROS facilities to build the framework prototype. We also use some high-performance plugin-based mechanism to optimize the bottom of the framework. One may use the framework-provided API conveniently to construct single-robot and multi-robot applications with all ROS resources available. This work is demonstrated by three application cases including an autonomous roving robot, a security patrol robot and multiple patrol robots. They are constructed and tested in both the simulated and the real environment. The experimental results validate the usability and availability of ALLIANCE-ROS.

Deep reinforcement learning (DRL) has greatly improved the intelligence of AI in recent years and the community has proposed several common software to facilitate the development of DRL. However, in robotics the utility of common DRL software is limited and the development is time-consuming due to the complexity of various robot software. In this paper, we propose a software engineering approach leveraging modularity to facilitate robot DRL development. The platform decouples learning environment into task, simulator and hierarchical robot modules, which in turn enables diverse environment generation using existing modules as building blocks, regardless of the underlying robot software details. Experimental results show that our platform provides composable environment building, introduces high module reuse and efficiently facilitates robot DRL.

Deep reinforcement learning is making advances in robotics with the platforms of realistic environment simulation. However, as shown in this paper, the realistic simulation introduces vast time cost which is the bottleneck of the learning procedure. To solve this problem generally, we propose a parallel reinforcement learning platform which follows the master-slave principle and integrates learning programs with multiple distributedrobot simulators. The platform is intrinsically scalable and requires no modification to existing serially designed learning environments or algorithms. Experimental results demonstrate that our platform significantly accelerates the learning progress of robots, in direct proportion to the parallel scale. The parallelism also brings richer exploration and sampling, enhancing the performance of deep reinforcement learning algorithms compared with existing serial platforms.

The agent-based subsumption model is widely acknowledged as the control systems for mobile robots. In this model, incremental layers can be stacked together by inhibitors and suppressors, which increasingly leads to complex and coordinate behaviors. ROS (Robot Operating System) is an open source robot software platform and gradually becoming the de facto standard for robot applications. There are abundant reusable function units in ROS, which can be coupled by a distributed messaging mechanism. This paper describes a template based on ROS for implementing the sub-sumption model, such that one can develop control systems for mobile robots by leveraging ROS-provided software resources. The publish/subscribe messaging mechanism in ROS is used to connect the loosely coupled modules of each layer. The behaviors inside a module are formalized as easy-to-use ROS-based finite state machines. The inhibitors and suppressors among layers are represented as ROS nodes and implemented as templates, which can be easily instantiated. And the work is demonstrated by two experiments. First, a three-layer autonomous wander robot, which is originally designed by Brooks, is reproduced in the ROS simulating environment. Second, a six-layer security patrol robot application, which is controlled by the subsumption model, is constructed and tested in the real world.