Text this: Cooperative multi-agent reinforcement learning in sparse-reward, partially observable 3d environments with curriculum-transfer learning