I am a Ph. D. candidate in Department of Electrical and Computer Engineering, National University of Singapore, and Institute for Infocomm Research, since 2002, under the supervision of Dr. Winston Khoon Guan Seah, and Dr. Marcelo H. Ang Jr.. My research interests include practical multi-robot surveillance, hybrid robot/sensor networks, robot/sensor localization in unstructured environment, multi-robot/agent learning, etc.

I received my B. Eng. degree from Department of Automation, Tsinghua University, Beijing, China, in 2001.

Surveillance is a wide research topic including many research fields such as exploration, target searching, target tracking, localization, etc. To improve the traditional surveillance techniques which rely on static sensory devices and centralized control architecture, mobile robots are applied to extend the observation of the environment, accelerate the searching of targets, enhance the performance of target tracking and monitoring, and increase the reliability and robustness of the entire surveillance system. In my study, a series of distributed multi-robot approaches are proposed for practical surveillance. The approaches cover exploration, target searching, target tracking, and localization problems.

Exploration and target searching: Some algorithms are proposed like the potential field-based exploration, swarm intelligence exploration, landmark-based exploration, hop-count gradient orientated searching, and cluster based searching. These methodologies can improve the observance of the environment and shorten the time spent for searching.

Target tracking: An artificial potential field-based intelligent tracking algorithm is proposed to let multiple moving targets be tracked by cooperative robots. This algorithm can assign the targets to the most suitable robots autonomously in a distribute manner.

Localization: Unlike traditional robot localization algorithms, my research interests in multi-robot localization are in the context of wireless sensor networks. In recent years, the hop-count-based localization scheme is proposed and studied for infrastructure-less wireless ad hoc networks. This localization scheme is simple and scalable; however, if the nodes are sparsely distributed in the networks, the location estimation may be quite inaccurate. To address this problem, a distributed auction-based task allocation mechanism is proposed to let robots proactively change the node density of the networks to improve localization precision. Through simulations, the system is found to produce more accurate and consistent results as compared to other approaches.

Above work has been tested by both simulations and experiments. More information can be found in Photo & Video.

For wireless ad-hoc sensor networks, how to enhance the network connectivity is one of the most important research issues. There indeed exist many schemes that already attempt to enhance network connectivity and information delivery without the use of "intelligent mobility", i.e., robots. However, most of these schemes are only able to minimize or delay the occurrence of network partitions. They are unable to repair the network disconnections arising from network partitions, whereby two groups of nodes are physically too far apart to send or receive wireless signals. To address the problems, we can change the network topology through static topology control (e.g., increasing the communication range of the nodes). However, static topology control is strictly limited and confined to the physical locations of the sensors. To have better network performance, controlled mobility should be applied to improve the sensor placements and thus enhance the network connectivity and information delivery.

With regard to above problem, a multi-robot system is designed to enhance the network connectivity by mobile robots. In this approach, the robots can explore and search in the environment cooperatively to find the network partitions, and deploy new sensor nodes in the critical regions to "bridge" the communication gaps.

Multi-robot mobility enhanced network connectivity: To find the communications gaps and "bridge" such partitions, the robots need to patrol the environment cooperatively. The cooperative exploration and target searching techniques can be applied to achieve efficient searching of the communication gaps. Coverage-centric and target-centric algorithms have been proposed as the potential field-based exploration, swarm intelligence exploration, landmark-based exploration, hop-count gradient orientated searching, and cluster-based searching. Simulation and experiment tests show that this robotics system can greatly improve the network data delivery (throughput) than other techniques.

Multi-robot mobility enhanced localization: For ad-hoc sensor networks, the hop-count-based localization scheme is usually used due to its simplicity and scalability. However, if the connectivity of the network is low, the location estimation can be quite inaccurate. To solve this problem, multiple cooperative robots are introduced to the network to search for and "bridge" the "non-critical" regions which may lead to excessive and inaccurate hop count.

Above work has been tested by both simulations and experiments. More information can be found in Photo & Video.

For multi-robot systems, one of the key research problems is to achieve cooperation among robots by decentralized (distributed) control methodology. However, the development and coding work for task level distributed controller is undesirable and sometimes extremely difficult, especially when the mission is complex and the robots are heterogeneous. To solve this problem, two kinds of machine learning techniques are proposed for multi-robot systems to let robots learn how to cooperate without the need for human design or coding.

Learning based on behavior-based control networks: In this approach, a reinforcement learning controller integrating reinforcement learning and behavior-based control networks is proposed to address the limitation of traditional reinforcement learning of discrete and finite behaviors. In addition, a distributed learning coordination algorithm is introduced to solve the problems in multi-robot concurrent learning. The learning controller and learning control algorithm are applied to multi-robot concurrent tracking of multiple moving targets.

Fuzzy reinforcement learning: In this approach, the discrete/finite states and actions are fuzzified as continuous/infinite fuzzy states and actions, and then the reinforcement learning controller learns based on these fuzzy states and actions to find the optimal fuzzy rules. In addition, motivated and inspired by human behaviors, a distributed learning coordination algorithm is derived to coordinate concurrent learning processes. The proposed learning controller and the learning control algorithm are applied to multi-robot concurrent tracking of multiple moving targets.

Above work has been tested by both simulations and experiments. More information can be found in Photo & Video.

In July 2002, I joined the robotics group. This group is headed by Dr. Marcelo H. Ang Jr.. In this group, my research focuses on the multi-robot systems including exploration, target searching, target tracking, and multi-robot localization. The key is to design/propose the methodologies to achieve cooperation among heterogeneous robots in a distributed and robust manner. For this purpose, some heuristics on the traditional behavior-based control techniques are proposed and applied. More important, some machine learning approaches are designed to enable the robots to learn how to cooperate. The learning algorithms are generic and robust such that they can be applied to most real-world multi-robot systems.

The work for robotics group has been tested by both simulations and experiments. More information can be found in Photo & Video.

In Jan 2003, I joined TARANTULAS project in Institute for Infocomm Research. This project is headed by Dr. Winston Khoon Guan Seah.

The full name of TARANTULAS is The All-teRrain Advanced NeTwork of Ubiquitous mobiLe Asynchronous Systems project. The aim of this project is to develop a system of networked heterogeneous mobile robots and static sensors to work together in any terrain to complete a given task, such as crisis management, environmental monitoring, inventory control, surveillance, etc.

This project involves the challenging task of integrating technologies from three key research areas: robotics, ad hoc networking, and localization and positioning.

Robotics: to design cooperative multi-robot system for

• Improved localization and communication in infrastructure-less wireless Ad-hoc Networks
• Better accomplishment of mission, such as crisis management, environmental monitoring, inventory control, surveillance, etc

Ad-Hoc Networking: to develop the suitable routing and MAC protocols that can improve the ad-hoc network for

• Timely information delivery
• Reliable information delivery
• Energy conservation

Localization and Positioning: to improve the performance of infrastructure-less hop-count-based localization in the presence of non-uniform node distribution and sparse node density through

• Mobility-enhancement
• Density-awareness
• Ultrasonic-enhancement

My work is mainly in the robotics part of TARANTULAS as the following:

• Design/assemble/maintenance the robot systems: AI-MRKIT, Koala, PPRK, etc
• Interface the robot with the communication system: MicaZ
• Conduct realistic simulations on both robotics and communications: Player/Stage + semaphore + GloMoSim (or QualNet)
• Conduct experimental tests on the whole system

The work for TARANTULAS project has been tested by both simulations and experiments. More information can be found in Photo & Video.