------------------
Change: In Python agents, the OnVideo callback now takes a PIL (Python Imaging Library) Image, instead of a raw buffer of bytes. See the sample scripts for usage hints.
Change: Needs 'requests': pip install requests
Change: You need an Experiment Studio human account created in order to log into Human Action.
Change: OutputUri is now required in MissionInit nodes
Add: Parameterisation added to the grid observation producer (the one that produces the hull of blocks around the player).
Add: Parameterisation added to the discrete world builder (produces a flat maze-like grid of stained glass over lava).
New MissionHandlers available (see the md files for details):
WorldBuilder: "WorldFromDefaultGame" - basically loads a default world.
ObservationProducers: "ObservationFromHotBar" and "ObservationFromFullInventory"
CommandHandler: "CommandForInventoryToHotBar" - allows basic manipulation of inventory.
Fixes to Discrete MissionHandlers: Discrete.RewardForGridWalking should now give a negative reward for attempting to move, even if the move was unsuccessful (eg the agent tried to walk into a wall).
Stabilisation - hopefully the Mod can now cope with thousands of missions without crashing, hanging, or otherwise doing weird things.