A new algorithm is being developed by researchers so as to help robots plan their actions in a proper way, when it comes to complex environments.

According to Brown University, they are looking forward to make these robots more useful in real life, however they are being developed in the virtual world of Minecraft. The study could help develop better action-planning skills for robots, deal with the phenomenon in which there are several choices named as ‘state-space explosion’. It ‘boggles’ the robot's intuition.

According to Stefanie Tellex, assistant professor of computer science at Brown, it’s a quite complicated problem. According to Tellex, "We want robots that have capabilities to do all kinds of different things, but then the space of possible actions becomes enormous. We don't want to limit the robot's capabilities, so we have to find ways to shrink the search space".

The new algorithm makes use of ‘goal-based action priors’; these are sets of actions in a given space that probably help a robot reach its goal. An operator can apply these priors, or these could be learnt through trial and error. The effectiveness of this algorithm was tested in the game Minecraft.

Tellex said that Minecraft is a quite good model for several robot problems and there are many possible actions that could be done by someone playing this game; it’s quite cheap and easy to gather a ton of training data, whereas it’s not easy to do that in the real world.

Small domains were created within the game by researchers and a task within that environment was solved using the algorithm.