We investigate the existence of computationally inexpensive first and second order statistics that uniquely describe grass for application in an autonomous lawnmower. We then segment images based on these statistics to determine locations of driveable terrain in an image. Tight statistical clustering of illuminated grass versus artificial texture suggests that this method is sufficient for identifying driveable terrain for an autonomous lawnmower.

Distributed denial-of-service (DDoS) attack is a serious threat in Internet. We propose a bit marking concept to identify and drop the DDoS attack packets. Bit marking is a variation of the packet marking technique that modifies packet headers at each router. However instead of storing the router information in the packets, bit marking alters one or more bits in the marking field. The bit marking process discussed in this paper is performed to all the packets and at all the routers along the path; hence it is called deterministic bit marking (DBM). DBM creates a common path signature for all the packets originating from the same location upon arriving at a destination. Since different source networks generate virtually unique path signatures, DBM makes it possible to isolate and discard DDoS attack traffic. From the Internet topology of autonomous systems we observe that the source networks are quite uniformly distributed over the path signature space. In our simulation over 99% of the attack traffic is blocked using DBM while up to 99% of the legitimate traffic passes. DBM can also be used for source traceback using reverse bit marking. DBM can be independently deployed for each ISP and the DBM-based networks can be protected from the attacks coming from nonDBM networks.

We present a hardware solution that can reliably block most of the malicious TCP traffic at the edge routers while passing the legitimate TCP traffic during a distributed denial-of-service (DDoS) attack on the Internet. By allocating bandwidths separately for TCP, the TCP portion of the bandwidth can be protected. In a simulation study, the filter successfully blocked 99.9% of the attack traffic while legitimate traffic showed nearly identical performance as in the non-attacked condition. This filtering is transparent to the hosts or routers and a filtering device can be easily attached to router ports.

Summarizes results of a five-year, multi-disciplinary, university-industry collaborative effort investigating design issues in agile manufacturing. The focus of the project is specifically on light mechanical assembly, with the demand that new assembly tasks be implementable quickly, economically, and effectively. Key to achieving these goals is the ease of equipment and software reuse. Design choices for both hardware and software must strike a balance between the inflexibility of special-purpose designs and the impracticality of overly general designs. We review both our physical and software design choices and make recommendations for the design of agile manufacturing systems.

The flexibility required of an agile manufacturing system must be achieved largely through computer software. The system's control software must be adaptable to new products and to new system components without becoming unreliable or difficult to maintain. This requires designing the software specifically to facilitate future changes. As part of the Agile Manufacturing Project at Case Western Reserve University, we have developed a software architecture for control of an agile manufacturing workcell, and we have demonstrated its flexibility with rapid changeover and introduction of new products. In this paper, we describe the requirements for agile manufacturing software and how our software architecture addresses them.

An agile workcell has been developed for light mechanical assembly in collaboration with industrial sponsors. The workcell includes multiple Adept robots, a Bosch conveyor system, multiple flexible parts feeders at each robot's workstation, CCD cameras for parts feeding and hardware registration, and a dual VMEbus control system. Our flexible pairs feeder design uses multiple conveyors to singulate the parts and machine vision to locate them. Specialized hardware is encapsulated on modular grippers and modular worktables which can be quickly interchanged for assembly of different products. Object-oriented software (C++) running under VxWorks, a real-time operating system, is used for workcell control. An agile software architecture was developed for rapid introduction of new assemblies through code re-use. A simulation of the workcell was developed so that controller software could be written and tested off-line, enabling the rapid introduction of new products.

It has been recognized that lead time and costs can be reduced by representing product designs in terms of manufacturing features such as surface finish, accuracy, or machining operations. Usually, a design environment supporting such representations is referred to as a primitive-based CAD system. This paper describes an object recognition system for range images which exploits the primitive-based CAD database information to reduce set up time and accelerate object identification. The heart of our work resides in modeling objects as collections of instantiated machining manufacturing primitives. This permits the implementation of a recognition-by-components approach using surface topology analysis. Thus, the main recognition task is broken down into the less stringent ones of primitive identification and instantiation, followed by the actual object model indexing. The result is faster recognition as well as automatic generation of object models from the CAD output databases.

This paper introduces a design for agile manufacturing workcells intended for light mechanical assembly of products made from similar components (i.e. parts families). We define agile manufacturing as the ability to accomplish rapid changeover from the assembly of one product to the assembly of another product. Rapid hardware changeover is made possible through the use of robots, flexible part feeders, modular grippers and modular assembly hardware. The flexible feeders rely on belt feeding and binary computer vision for Dose estimation. This has a distinct advantage over non-flexible feeding schemes such as bowl feeders which require considerable adjustment to changeover from one part to another. Rapid software changeover is being facilitated by the use of a real-time, object-oriented software environment, modular software, graphical simulations for off-line software development, and an innovative dual VMEbus controller architecture. These agile features permit new products to be introduced with minimal downtime and system reconfiguration.

A technique for evaluating the learning capability and optimizing the feature space of a class of higher-order neural networks is presented. It is shown that supervised learning can be posed as an optimization problem in which inequality constraints are used to code the information contained in the training patterns and to specify the degree of accuracy expected from the neural network. The approach establishes: (a) whether the structure of the network can effectively learn the training patterns and, if it can, a connectivity which corresponds to satisfactorily learning; (b) those features which can be suppressed from the definition of the feature space without deteriorating performance; and (c) if the structure is not appropriate for learning the training patterns, the minimum set of patterns which cannot be learned. The technique is tested with two examples and results are discussed.

A 3D object recognition system for parts designed with a form feature CAD is presented. Objects are modelled as organized compositions of instantiated form primitives. Because of this, the burden of part recognition is shifted from matching the scene to the part database to matching the scene to the primitive database, which is usually a significantly smaller database. In this paper we briefly review our approach to primitive and object representation, the architecture of the system, hypotheses generation, and primitive indexing. Results obtained wish both synthetic and actual range images are presented.>

A new method, called feature extraction by demands (FED), for generating an object description concurrently at different feature levels will be described. An object is described in terms of features which include points, surface patches, edges, corners, and surfaces. These features form a feature space which is the base used to decompose the feature extraction process into different levels. FED provides a method to generate partial descriptions about objects from partially processed range data at different feature levels. The partial descriptions become a feed-back to guide the feature extraction process to extract more detailed information from interesting areas which can then be used to refine the object description. Regions which are not perceived to contain useful infomation will be ignored in further processing. As a more complete object description is generated, FED converges from bottom-up image processing to top-down hypotheses verification to generate complete hierarchical object descriptions.