Yizhen Zhang

Particle swarm optimization for unsupervised robotic learning

We explore using particle swarm optimization on problems with noisy performance evaluation, focusing on unsupervised robotic learning. We adapt a technique of overcoming noise used in genetic algorithms for use with particle swarm optimization, and evaluate the performance of both the original algorithm and the noise-resistant method for several numerical problems with added noise, as well as unsupervised learning of obstacle avoidance using one or more robots.

A new threat assessment measure for collision avoidance systems

Collision avoidance systems (CAS) are an emerging automotive safety technology that assists drivers in avoiding collisions. They monitor the dynamic traffic and vehicle information in real time and assess the current threat level to decide if warnings should be issued to the driver or automatic braking is needed to avoid collisions. Several measures have previously been defined for threat assessment with various warning and overriding algorithms proposed. A new threat assessment measure, time-to-last-second-braking (Tlsb) is proposed in this paper and its advantages over previous measures are discussed. It directly quantifies the danger or threat level of the current dynamic situation objectively as well as assesses the urgency level for the required evasive action, e.g., braking. It is in agreement with human natural judgment of the urgency and severity of threats. Furthermore, new warning and overriding criteria are proposed based on the new Tlsb measure, providing more appropriate warning and more effective override timing

Evolving Engineering Design Trade-Offs

In the engineering design process, it is important to know the engineering trade-offs achievable under various design preferences and strategies. In this paper, a family of engineering design trade-offs are evolved from an automatic design synthesis methodology based on evolutionary computation. The complete Pareto optima frontier can be evolved by a consideration of fitness function that aggregates the weighted fuzzy design preferences under different trade-off strategies. An initial case study concerned with the configuration of a collective sensory system is presented and discussed, along with preliminary results obtained from simulations under a specific scenario. The results indicate that the approach can be useful for designers to solve complex engineering problems.Copyright

A realistic simulator for the design and evaluation of intelligent vehicles

The number of vehicles hitting the road each day is rapidly increasing, and several problems, such as traffic congestion or driver safety, can no longer be solved in the same fashion as before. Intelligent transportation systems could potentially solve part of these problems, but prototyping, designing and testing cooperative smart vehicles is a cumbersome task. This paper presents a realistic simulator, capable of operating both at microscopic and sub-microscopic level, where intelligent vehicles can be designed and analyzed with a pragmatic approach. A number of advances in robotics have already been transferred to vehicular technology, with a potential increase of this trend into the future. Here, we develop a plugin for a well-established robotics simulator (Webots), in order to reinforce at the virtual level this cross-fertilization between the two areas and create a baseline for realistic studies of future solutions in real intelligent vehicles.

Evolving Neural Controllers for Collective Robotic Inspection

In this paper, an automatic synthesis methodology based on evolutionary computation is applied to evolve neural controllers for a homogeneous team of miniature autonomous mobile robots. Both feed-forward and recurrent neural networks can be evolved with fixed or variable network topologies. The efficacy of the evolutionary methodology is demonstrated in the framework of a realistic case study on collective robotic inspection of regular structures, where the robots are only equipped with limited local on-board sensing and actuating capabilities. The neural controller solutions generated during evolutions are evaluated in a sensorbased embodied simulation environment with realistic noise. It is shown that the evolutionary algorithms are able to successfully synthesize a variety of novel neural controllers that could achieve performances comparable to a carefully hand-tuned, rule-based controller in terms of both average performance and robustness to noise.

Evolution of sensory configurations for intelligent vehicles

An evolutionary design synthesis methodology was introduced with special concern for the design and optimization of distributed embodied systems. Its efficacy was validated in a case study on the design of collective sensory configurations for intelligent vehicles. Candidate sensory configurations were tested in sample traffic scenarios simulated in an embodied and sensor-based simulator, and in more abstracted and computationally efficient evaluation tests. Sample results evolved under different design preferences are presented, including approximate Pareto fronts representing the engineering design trade-offs characterizing the problem investigated in the case study.