This paper presents an new, proximate-optimal solution to the path-constrained time-parameterization problem. This new algorithm has three distinguishing features: First, the run-time worst-case complexity of the proximate time-optimal algorithm is linear with respect to path-length and it is shown to be more efficient than any other truly time-optimal algorithm. Second, for a given robotic system, the algorithm's running-time is predictable as a function of the length of the path (allowing its use in combination with time-aware planners). Third, the algorithm easily supports the modification of on-going trajectories. The algorithm has been extensively tested and is operational in a number of robotic systems including a dual-arm workcell, an underwater robotic system, and the Marsokhod Rover vehicle. Experimental results presented illustrate the online use of the algorithm with a path planner to allow capture and delivery of objects from a moving conveyor belt.

This paper introduces a design technique for complex robotic systems called interfaces-first design. Interfaces-first design develops information interfaces based on the characteristics of information flow in the system, and then builds subsystem interfaces from combinations of these information interfaces. This technique is applied to a dual-arm workcell combining a graphical user interface, an on-line motion planner, real-time vision, and an on-line simulator. The system is capable of performing object acquisition from a moving conveyor belt and carrying out simple assemblies, without the benefit of pre-planned schedules nor mechanical fixturing. The information characteristics of this system are analyzed, and divided into three interfaces: world state, task command, and motion commands. Detailed descriptions of the resulting interfaces are provided. The paper concludes with experimental results from the workcell. Both single-arm and dual-arm actions are discussed.

Real-time system software is notoriously hard to share and reuse. This paper walks through the methodology and application of ControlShell, a component-based programming system real-time system software development. ControlShell combines graphical system-building tools, an execution-time configuration manager, a real-time matrix package, and an object name service into an integrated development environment. It targets complex systems that require on-line reconfiguration and strategic control. ControlShell takes advantage of functional object hierarchies to enable code sharing and reuse. It gains flexibility by supporting easy interconnectivity of these objects. It features a unique configuration control system for changing operating modes. The paper concludes by examining the application of this framework to a dual-arm robotic workcell that is able to pick objects from a moving conveyor and perform simple assemblies,.

The network data delivery service (NDDS) is a novel network data-sharing system. NDDS builds on the model of information producers (sources) and consumers (sinks). Producers generate data at their own discretion, unaware of prospective consumers. Consumers "subscribe" to data-updates without concern for who is producing them. The routing protocol is connectionless and nearly "stateless", thus network reconfigurations, node failures, etc. are handled naturally. This scheme is particularly effective in systems (such as distributed control systems) where information is of a repetitive nature. NDDS provides support for multiple producers, reliable data-delivery, consumer update guarantees, notifications vs. polling for updates, dynamic binding of producers and consumers, distributed queries, and user-defined data types. NDDS is integrated into the ControlShell real-time framework and is being used in several robotic applications as an effective means of information sharing between sensor systems, robot controllers, planners, graphical user interfaces, and simulators.>

A high-level controller for a SMALL workcell (Sawyer-motor assembly workcell) is presented. It allows multiple independent process to share the robot modules and workspace of a single workcell, by planning and scheduling collision-free paths for module motions in real time. Planning may involve moving around other modules, stationary or in motion, waiting for motions to complete, or displacing obstructing idle modules. The controller handles any number of robots in simultaneous motion along paths consisting of multiple rectilinear segments. Collision avoidance and path-planning are addressed in two dimensions, with extensions to three dimensions briefly discussed. Design considerations for the algorithms involve throughput, protection, fairness, and starvation considerations. Capabilities to allow independent processes to cooperate without interference are provided.>