• View
• Edit
• History
• Print

System

Systems

These systems are the code embodiment of the various architectures conceived. As such, they are likely not perfect models but instead are pragmatic compromises. But they run! Or at least they used to, to some extent.

Three Level

These systems subscribe (by varying degrees) to the decomposition of robotic systems into 3 distinct layers. How the distinctions are drawn, and what exactly each level is responsible for varies. The three levels are (from highest to lowest): deliberative, executes state space searches to determine a set of actions to perform; sequencer (or executive), takes the goals decided on by the deliberative layer and translate them into a sequence of commands; reactive, responsible for control loops and sensor / actuator interaction.

Of these, 3T is likely the most prototypical. It utilizes RAP as its sequencer and builds upon the work of several other systems such as ATLANTIS. CLARAty was intended to succeed 3T, but does not appear to have been very successful.

Graph

The graph-based systems shift focus to allowing users to create a layer of interacting components. This layer is supported by a collection of libraries and drivers.

The earliest of the graph-based systems appears to be CARMEN, which grew out of the 3T effort. Player formalized the idea of components. While these components started out as drivers in the conventional sense (interfacing with and abstracting hardware), the ability to connect the output from one driver to the input of another facilitated a major shift. Drivers began to become more and more sophisticated: from filtering sensor data to aggregating several sources to controlling basic obstacle avoidance to (finally) complex planning and control algorithms.

The popular ROS is the successor to Player. YARP is similar, but emphasizes a light-weight, less restrictive libraries only approach.

Distributed

These systems are designed to run across several physical machines. Most often considered for more resource constrained platforms, these systems are typically very light weight. Focus is on achieving networking goals while maintaining a small footprint. Events are a common abstraction.

TinyOS is, as one would expect, very tiny. It is targeted at sensor networks. ASEBA was originally designed for a heterogeneous swarm, but has seemingly morphed into a suite of tools for an educational line of robots.

Model Based

The fault tolerance and robustness demanded by space missions lead to a more formal approach to robotic systems: model based. In these systems, the planner is provided with a formal model and is able to plan even in the face of actuation errors and sensor errors. While powerful, the resource demands and time complexity are too great for many domains.

Titan and Remote Agent both fall into this category.

Other

Thus far neglected in this list are Wombat; a system developed for the DARPA Grand Challenges by MITRE, ALLIANCE a behavioral approach to multi-robot systems, and of course whatever may have made it under the Other category.