A new research project from MIT is trying to tackle the state of uncertainty in robotics collaboration.
The research team is dealing with a type of robot automation called Decentralized Partially Observable Markov Decision Processes (Dec-POMDPS) – a set of algorithms used to define the way a system behaves. The MIT researchers are looking to solve the uncertainty that accompanies robotic synchronisation, for example controlling a compelx system of delivery drones prone to collisions.
The study, titled ‘Planning with Macro-Actions in Decentralized POMDPs‘ [PDF], explains the always-present amount of error in the sensor data which each drone uses to stick to its path. When complicated tracks need to be taken, the risk of error can make a journey very difficult to plan.
The published paper proves that Dec-POMDPS can be employed to combine existing robotic control networks to achieve tasks in collaboration.
Using remote-controlled helicopters and simulating a drone delivery service, a test was set up to investigate the accuracy of a Dec-POMDPS algorithm across a number of base stations and delivery locations. The helicopters would need to travel across each other’s paths to deliver the packages without crashing.
Offline plotting was required before the launch, marking an approximate route before the algorithms took over.
The scene was then split into two graphs by the algorithms – one generating a set of potential macro-actions, and the other generating transitions between those macro-actions in line with observations from all the drones within the system.
The planning algorithm pairs a value to each macro-action which are then presented on a graph to show the probability that a drone should carry out a particular action at a particular time. This process was repeated for each macro-action until all of the drones had reached their final destinations safely.
As these collaborative systems based on algorithms of uncertainty develop, some have commented upon the scope for malicious actors to intrude and automate potentially disastrous incidents.