REWARDs

REWARDs are scalar real values associated with each LOCATION in the environment. By definition, every LOCATION must have a REWARD, although most of those REWARDs may be zero. When the AGENT enters a LOCATION containing a non-zero REWARD, it receives that REWARD as feedback: positive REWARDS are desirable and the AGENT is to learn to seek them out; negative REWARDs are detrimental and the AGENT is to learn to avoid them.

The WORLD SIMULATOR is responsible for tracking REWARDs and deciding when to issue REWARD feedback to the AGENT, based on the AGENT's current STATE.