Context Transfer in Reinforcement Learning Using Action-Value Functions

Figure 1

A grid as a farm with three crops and three harvesting robots. Robot 1: Sensor modules: GPS, color&weight sensor, , : column number, : row number, : Red, : Green, : Yellow, : Light, : Heavy, 0: Nothing, , : Move North, : Move South, : Move East, : Move West, 0: Nothing, : Pickup, : Dropoff. Robot 2: Sensor modules: GPS, Compass, B&W camera, , are the same as robot 1, : direction, : Small Globe, : Rod, : Big Globe, 0: Nothing, , : Move Forward, : Move Backward, : Turn left, : Turn Right, : Turn left & , : Turn right & , 0: Nothing, : Pickup, : Dropoff. Robot 3: Sensor modules: beam’s signal distance indicator, Compass, color & weight sensor, , : 1-norm distance to beam , is the same as robot 2 and as robot 1, .