Make it so that POUCT/POMCP planning result could be deterministic. This requires seeding both sampling states from the belief and sampling from generative models. A random generator or a seed number should be provided when creating the planner and passed into the agent. This would imply quite a lot of changes as each domain would have to update the sample methods of its models to be using the passed-in random generator. By default, no seed is provided and the behavior should be random (non-deterministic).
Make it so that POUCT/POMCP planning result could be deterministic. This requires seeding both sampling states from the belief and sampling from generative models. A random generator or a seed number should be provided when creating the planner and passed into the agent. This would imply quite a lot of changes as each domain would have to update the sample methods of its models to be using the passed-in random generator. By default, no seed is provided and the behavior should be random (non-deterministic).