Robust distributed computing and sensing algorithm

R.R. Brooks, S.S. Iyengar
1996 Computer  
I Sensors that supply data to computer systems are inherently unreliable. When sensors are distributed, reliability is further compromised. How can a system tell good sensor data from faulty? A hybrid algorithm combines proposed iolutions to address the problem. 0018-9162/96/$5 00 0 1996 IEEE ur modern world contains many automated systems that must interact with changing environments. Because these environ-0 ments cannot be predetermined, the systems rely on sensors to provide them with the
more » ... ormation they need to perform their tasks. Sensors providing data for control systems are the unenviable interface between computer systems and the real world. Programming automated control systems is difficult because sensors have limited accuracy, and the readings they return are frequently corrupted by noise. To avoid systems being vulnerable to a single component failure, it is reasonable to use several sensors redundantly. For example, an automatic tracking system could use different kinds of sensors (radar, infrared, microwave) that are not vulnerable to the same kinds of interference. Redundancy presents a new problem to system designers because the system will receive several readings that are either partially or entirely in error. It must decide which components are faulty, as well as how to interpret at least partially contradictory readings. To improve sensor-system reliability, researchers have actively studied the practical problem of combining, orfusing, the data from many independent sensors into one reliable sensor reading. When integrating sensor readings, robustness and reliability are crucial properties. It is increasingly obvious that sensor integration, which must include some type of fusion, is necessary to automate numerous critical systems.l Redundant sensors in an automated control system form one type of distributed system. A key advantage of distributed computing is that it adds a new dimension of integrity to computing. Compuiations made by a network of independent processors are insensitive to a single hardware failure. Instead, the concerns in a distributed system are determining how many component failures a network can tolerate how the network separates the output from correctly functioning and still be reliable and machines from that of defective machines. The central question is, how can an automated system be certain to make the correct decision in the presence of faulty data? Much depends on the system's accuracy--the distance between its results and the desired results-and on the system's precision-the size of the value range it returns. To solve the problem algorithmically, we basically have sensorfusion and Byzantine agreement. Danny Dolev2 presented one Byzantine agreement algorithm to solve the Byzantine generals problem posed by Leslie Lamport and colleague^.^ The Byzantine generals problem presupposes a distributed decision-making process in which some parti'cipants not only
doi:10.1109/2.507632 fatcat:lbo33nkjcfh2plnfpy72qkwvwm