ErikPolzin / PackageCollector

Package collection via reinforcement learning

Apply reinforcement learning to an agent in a 13x13 four-room bounded environment. Runs under the following scenarios:

• 1: Single package
• 2: Multiple packages
• 3: Multiple packages with ordered collection

Build & launch instructions

Run make inside the extracted folder. Pip will install its dependencies inside a virtual environment

Command-line interface

usage: ExecutionSkeleton [-h] [-stochastic] [-learning-rate LEARNING_RATE] [-discount-rate DISCOUNT_RATE]
                         [-epochs EPOCHS] [-test] [-save SAVE]
                         scenario

Trains the FourRooms Agent in a given scenario

positional arguments:
  scenario

options:
  -h, --help            show this help message and exit
  -stochastic
  -learning-rate LEARNING_RATE
  -discount-rate DISCOUNT_RATE
  -epochs EPOCHS
  -test
  -save SAVE

Files

• FourRooms.py: Environment framework
• ExecutionSkeleton.py: Agent definition and runners
• Scenario1.py: Run the simple scenario
• Scenario2.py: Run the multi scenario
• Scenario3.py: Run the rgb scenario