The MineRL 2020 Competitors aims to foster the development of algorithms which can successfully leverage human demonstrations to drastically decrease the amount of samples essential to resolve complicated, hierarchical, and sparse environments.
To that finish, individuals will contend to develop methods that can get hold of a diamond in Minecraft from uncooked pixels employing only 8,000,000 samples from the MineRL simulator and 4 days of training on a one GPU machine. Individuals will be offered the MineRL-v0 dataset (web-site, paper), a large-scale selection of in excess of sixty million frames of human demonstrations, enabling them to make use of qualified trajectories to minimize their algorithm’s interactions with the Minecraft simulator. More detailed background on the opposition and its design and style can be found in the MineRL 2020: NeurIPS Competitors Proposal.
The undertaking of the opposition is solving the MineRLObtainDiamondVectorObf-v0 environment. In this environment, the agent starts in a random starting up location with no any things, and is tasked with obtaining a diamond. This undertaking can only be achieved by navigating the complicated merchandise hierarchy of Minecraft.
The agent gets a superior reward for obtaining a diamond as perfectly as smaller sized, auxiliary benefits for obtaining prerequisite things. In addition to the major environment, we give a amount of auxiliary environments. These consist of responsibilities which are either subtasks of ObtainDiamond or other responsibilities within just Minecraft.