Algorithm Developed That Creates Contingency Plans

[Guest_Jim_*]

'The best laid plans of mice and men often go awry' is a saying whose truth one can never fully escape from. This is why it is often a good idea to be ready with other plans for when things go wrong. For humans, developing and switching to a contingency plan is not necessarily that hard, and soon it may be less difficult for machines, thanks to researchers at MIT and the Australian National University.

For some time now planning algorithms have been used to help construct optimal schedules and control policies for very complex systems, and to help direct autonomous robots. When something goes wrong though, and the original plan becomes too risky, a contingency plan will be needed, but creating one has traditionally been quite difficult, for a machine. This is because the algorithms have to determine risks to different plans with probability distributions, which are not linear and thus somewhat challenging to work with. They can however be approximated with linear functions, and the approximations used in this new algorithm will only ever underestimate risk, so if it throws out a possible plan, it is guaranteed to be one that would be too risky to accept anyway. The acceptable level of riskiness is something the user can set too, so when the algorithm finds a plan is going to exceed the risk budget it will stop and move on to another string of choices.

With the different optimizations of this algorithm, such as the use of heuristics and the linear approximations, the hope is it can be used in realtime by machines. This would especially important for various space missions, when the robotic spacecraft has to carry out its mission but not endanger itself too much.

Source: MIT

Back to original news post