Home  |    Lab Home     |     Projects    |      People     |    Publications     |     Links    |    News    |

  Research Projects

     My major research interests include bio-inspired self-organizing systems, machine learning, artificial intelligence in bioinformatics, and real-time embedded systems.  Some of my recent research projects are summarized here.  For more details, please refer to my online selected publications by topic or publications by category.

 Bio-Inspired Self-Organizing Systems

Morphogenetic robotics is a new emerging field in developmental robotics, which is dedicated to the application of morphogenetic mechanisms to robotics.  Inspired by multi-cellular morphogenesis, currently, we have focused on two types of morphogenetic robotic systems: morphogenetic swarm robotic systems and morphogenetic modular robots.  

Morphogenetic Swarm Robotic Systems.  Biological systems, from macroscopic swarm systems of social insects to microscopic cellular systems, can generate robust and complex emerging behaviors through relative simple, local interactions subject to various kinds of uncertainties.  A major research focus of self-organizing swarm robotic systems is to create traits of biological systems such as self-reconfiguration, self-organization, and self-repair.  However, it is also well aware that it is difficult to predict the emerging behaviors from local interactions of the individual agents/elements neither is it trivial to design rules for local interactions to generate a designed global behavior.  The target of this project is to study the genetic and celullar mechanisms for generating robust emerging behaviors for collective systems.  Biological organisms have evolved to perform and survive in a world characterized by rapid changes, high uncertainty, infinite richness, and limited availability of information.  Morphogenesis is one fundamental biological process in developmental biology that guides an organism to develop its body plan. Multi-cellular morphogenesis is under the control of gene regulatory networks (GRNs), which are models of genes and gene interactions at the expression level. In this project, inspired by the biological morphogenesis, a set of distributed morphogenetic approaches have been developed for swarm robotic systems on various applications, such as complex pattern formation, boundary coverage, mobile targets entrapping, etc..  Here are some simulation/experimental videos we have conducted in this project.

Complex pattern formation using swarm robots (simulation)

Swarm robot pattern formation with a mobile obstacle

 

Self-organization of swarm robots (e-puck robots)

 

 

Self-organization of swarm robots for mobile targets entrapping

       Some more videos can be downloaded here.

• Forming a unit circle.  Video (506KB, duration 49 seconds)
• Forming arbitrary shape “NURBS”.    Video (211KB, duration 42 seconds)
• Forming an arbitrary shape “R”, then avoid a mobile obstacle.  Video (2.26MB, duration 45 seconds)
• Forming 3D shapes.  Video (1.42MB, duration 29 seconds)
• Arbitrary shape formation using multiple e-puck mobile robots. Video (17.7MB, duration 11seconds)
• Bird-flocking formation and mobile obstacle avoidance. Video (576KB)
• Boundary coverage simulation. Video (1.13MB)
• Bird-flocking using e-puck robots. Video (6.7MB)
• Local coordination with adaptive neighbors using e-puck robots. Video (14.4MB)

Morphogenetic Modular Robots. Self-reconfigurable modular robots are metamorphic systems that can autonomously change their logical or physical configurations (such as shapes, sizes, or formations), as well as their locomotion and manipulation, based on the mission and the environment. Although self-reconfiguration is believed to be the most important feature of modular robots, the ability to adapt their configuration autonomously based on onboard sensing information under environmental changes remains to be demonstrated.  Inspired by biological morphogenesis, recently, we have developed a hierarchical morphogenetic approach for lattice-based modular robots, which is inspired by the embryonic development of multi-cellular organisms.  Extensive simulation results have demonstrated the efficiency and robustness of this model under various dynamic environments. Based on the morphogenetic approaches, the proposed modular robots which can automatically self-reconfigure to appropriate patterns to adapt to dynamic environmental changes based on the onboard sensor information (most available models can only reconfigure the robots to some predefined patterns, which cannot handle environmental changes). Two different types of modular robots have been developed, Cross-Cube and Cross-Ball.  Here are some simulation/experimental videos we have conducted in this project.

Self-reconfiguration of modular robot (Cross-Cube) to adapt to environmental changes

Self-reconfiguration of modular robot (Cross-Ball) to various patterns

Self-reconfiguration of modular robot (Cross-Ball) to traverse a rough terrain

Some more videos can be downloaded here.

• Dynamically reconfigure modular robots to adapt to different environments. Video (10.2MB)

• Self-repair with system failures and traverse different pathways. Video (5.4MB)

  Bioinformatics

   Bioinformatics is the science of managing, mining, and interpreting information from biological sequences and structures. Previous applications of data mining and machine learning to bioinformatics include gene finding, protein function domain detection, function motif detection, disease diagnosis, disease prognosis, disease treatment optimization, protein and gene interaction network reconstruction, and data cleansing.  Some clustering algorithms have been proposed for these applications, such as K-means algorithms, Genetic Algorithm (GA), Self-Organizing Maps (SOM), Support Vector Machine (SVM), artificial Neural networks, and decision trees.   Even though most available classification methods are powerful in the information they produce, they may not be efficient or applicable to predict the outcomes of other clinic occasions.  Furthermore, the choice of classification algorithms can give different classifiers, which may or may not be biologically significant.  Therefore, in this project, we plan to develop a swarm intelligence based neural network technique for a generic analytical and prediction system, which can analyze and interpret the dataset, predict the results in a consistent way with known biological information about the clinic case at hand.  Eventually, this system will automatically provide an optimal treatment strategy for specific clinic occasion.  More specifically, we are applying this swarm intelligence based neural network based techniques on optimizing nerve regeneration, which is recently supported by a NIH grant. 

 Embodied Artificial Intelligence

The potential applications for robotic systems with general intelligence and adaptive behavior are virtually limitless.  The right cognitive architecture is necessary to create such intelligent robotic systems. However, most current cognitive architectures only focus on human cognitive abilities from different developmental environments specifically for different tasks.  We believe that the complex interactions that emerge in social contexts are necessary for agents to develop impressive cognitive capabilities. To this end, we are developing a Civilization-Inspired Vying Societies (CIVS) system, which is a novel evolutionary learning multi-agent system loosely inspired by the history of human civilization.  The system uses a bottom-up Artificial Life (Alife) approach with the goal of generating intelligent emergent behavior in agents and agent groups. CIVS proceeds from the idea that an agent will not develop complex, intelligent behavior in isolation; it must be part of a culture that supports such development.  CHARISMA cognitive architecture is being developed for social agents embodied in the CIVS system, and a dynamic knowledge representation, called SHYNE (Semantic HYper NEtwork), has been developed for embodied agents to learn knowledge/skills as well as share information with other agents.  This is an ongoing project, which aims to develop general cognitive intelligence for social agents that can handle various tasks automatically and adaptively through self-organizing and self-learning from interaction with environments and other agents.

    Here are two simulation scenarios we have developed for the CIVS system.

"Catch The Flag" Scenario Simulation

"Hudson River Estuary Port Security" Scenario Simulation

Machine Learning in Robot/Computer Vision

Visual Object Recognition, Learning and Tracking. Visual recognition and tracking of people and their behaviors is an important issue in many applications, such as service robots, surveillance systems, public security systems, virtual reality interfaces, and other government, commercial, and entertainment applications.  To tackle this problem, we propose a new object tracking algorithm that embeds swarming particles into generic particle filter framework (PSO-PF) to achieve more robustness and flexibility. The proposed algorithm is independent with specific objects and can be used for any free-selected object tracking. Some experimental results demonstrate efficiency and robustness of the algorithm.

     One main challenge in visual object recognition is to correctly represent feature distributions due to the significant data variances.  Biological evidence found in cognitive brain research and neuroscience suggests that the nervous system responsible for object recognition has distributed cortical structures containing both bottom-up and top-down pathways. Therefore, recently, we have developed a novel hierarchical neural network model for visual object recognition, which aims to explore the potentials of combining discriminative and generative data flows under the neural network architecture by fusing the bottom-up stimulus and top-down expectations. This bi-directional pathway based HNN model has been implemented on various huge dataset for different object recognition, and the experimental results demonstrated the efficiency and robustness of the proposed model.  

Behavior/Pattern Recognition and Learning. The objective of this project is to develop a vision system to automatically analyze and recognize human behaviors online from a sequence of videos.  The following three models have been developed to tackle this problem.

• Self-adaptive Hidden Markov Models (HMM). We proposed a self-adaptive HMM model, where if the unknown sequence cannot be classified into any trained HMMs, a new HMM will be generated and trained. In this manner, our models can be employed in on-line training to dynamically complete the high-level description of behaviors.

• Hierarchical Probabilistic Latent Model. A novel hierarchical probabilistic latent (HPL) model is proposed, which consists of four layers from bottom-up: spatiotemporal visual features, atomic pattern layer, latent topic layer, and behavior pattern layer. In this manner, the complicated human activities can be decomposed into low level features, atomic patterns, and latent topics, which are much better suited for the automatic understanding of human behaviors. Given a video sequence, both spatial and temporal interest points are extracted as the low level visual features, which are clustered into distributions of atomic patterns using hierarchical Bayesian networks (HBNs).  Then, the proposed hierarchical probabilistic latent model is applied to represent the behavior patterns and latent topics as distributions over atomic patterns.

• Bio-inspired Hierarchical Spiking Neural Networks. Inspired by the biological morphogenesis and biologically sound spiking neural networks, recently, we have proposed an evolving gene regulatory network based BCM (E-GRN-BCM) model, which is a spiking neural network combined with the biological cellular phenomena, and then apply this model to online human behavior recognition. In this new model, the neurons will have a spiking rather than continuous output behavior, and each neuron will maintain a recent history for the inclusion of heavily abstracted effects such as spike timing dependent plasticity, ion concentration, membrane potential, etc.  Some preliminary experimental results have demonstrated that the proposed model can achieve a higher behavior recognition rate from video sequences compared to other state-of-the-art methods.

Distributed Multi-Robot Systems

One of the main challenges for multi-robot systems is to create intelligent agents that can adapt their behaviors based on interaction with the environment and other agents, become more proficient in their tasks over time, and adapt to new situations as they occur.  Such ability is crucial for developing robots in human environments.  Swarm robots are often observed to display many of the attributes which are typical in collective intelligent system in general including robustness, adaptability, flexibility, and self-organization.  Typical problem domain for the study of swarm-based robotic systems including foraging, box-pushing, aggregation and segregation, formation forming, cooperative mapping, soccer tournaments, site preparation, sorting, and collective construction. The main challenge of these control architecture are scalability due to limited sensing and communication capability of swarm robots.  We have addressed this challenge from the following three different views. 

• Swarm-intelligence based approaches. By taking inspiration from the behaviors of social insects swarming, flocking, and herding, where social insect colonies are able to build sophisticated structures and regulate the activities of millions of individuals by endowing each individual with simple rules based on local perception, several novel swarm intelligence based approaches have been developed to control multi-robot systems in a distributed manner.  

• Heuristic rule based approaches. We developed an autonomous system consisting of cooperative mobile robots and Fiber Optic Sensors (FSs) for perimeter defense tasks. The system concept is that the FSs will sense perimeter intrusions and act as a cueing sensor to an ensemble of robots. These robots in turn engage the potential intruder, performing surveillance and/or neutralization of the intrusion. To minimize the intruder missing rate, some robots have to perform tracking of the intruders while others have to deploy themselves dynamically to cover the protected area. Therefore, a shame-level based dynamic task allocation algorithm is proposed for intruder tracking and allocation, and a gap-based algorithm is proposed for self-deployment of the remaining robots. Both algorithms are developed in a decentralized manner.  

• Multi-agent reinforcement learning. We developed a dynamic correlation matrix based   multi-Q learning (DCM-MultiQ) method for a distributed multi-robot system.  A novel dynamic correlation matrix is proposed, which not only handles each agent’s Q value in a multi-agent system, but also deals with the correlation among agents. Furthermore, a feedback matrix theory is applied to this dynamic correlation matrix so that the convergence of this DCM-MultiQ algorithm can be proved theoretically.

    Real-Time Embedded Systems

Reconfigurable System-on-Chip Architecture Design for Real-Time Systems:  Conventionally, all the tasks of a complex real-time system are implemented on a general processor in software only due to its great flexibility.  A powerful processor is needed to guarantee that hard deadlines always are met, even for very rare conjunctions of events.  Sometimes the sensors of a real time system are supposed to work in parallel and independently, it would be difficult to implement the parallel sensing in software. Furthermore, with multi-tasks, signals, events, it would be a challenging job for a software engineer to implement all of the tasks under real-time constraints especially with some computation-intensive information, say video images. On the other hand, the specific hardware can respond to external inputs more efficiently since the multiple hardware units can all operate in parallel, concurrent, and asynchronous, so that individual response times are much less variable and easier to guarantee, even as the number of tasks increases.  Parallel hardware is not so affected by issues such as task swapping, scheduling, interrupt service, and critical sections, which complicate real-time software solutions.  The expensive ASIC can fulfill the speed criteria. However, it is a complicated and expensive procedure if a slight change occurs.  The newest FPGA technology allows a designer to use a single reconfigurable platform to instantiate both the processors and the required logic units, which directly leads to feasibility of the   reconfigurable system-on-chip (rSoC) architecture. In this project, a unified agent-based hardware/software representation is proposed to simplify the hardware/software codesign procedure.  Currently, we are developing dynamic task allocation and scheduling algorithms for this rSoC platform.  This rSoC platform is desirable for those real-time systems which must be highly responsive to the dynamic environments. 

Development of SMARbot (Stevens Modular Autonomous Robot): Miniature robots have many advantages over their larger counterparts, such as low cost, low power, and easy to build a large scale team for complex tasks.  Heterogeneous multi miniature robots could provide powerful situation awareness capability due to different locomotion capabilities and sensor information. However, it would be expensive and time consuming to develop specific embedded system for different type of robots.  The goal of this project is to construct a simple, easy to use, embedded system framework that can generally meet or exceed the specifications of various robots, while maintaining a great deal of flexibility. In this project, we propose a generic modular embedded system architecture called SMARbot (Stevens Modular Autonomous Robot), which consists of a set of hardware and software modules that can be configured to construct various types of robot systems.  These modules include a high performance microprocessor, a reconfigurable hardware component, wireless communication, and diverse sensor and actuator interfaces. This miniature robot has been developed from scratch in our lab and currently has been used by course projects and other robotic research projects.

These research projects are partially sponsored by NSF, NIH, DoD, US Army, Honda Research Institute Europe, and Stevens Faculty Startup Funding.

Please send comments to: yan.meng@stevens.edu