Goal-directed navigation in humans and deep reinforcement learning agents relies on an adaptive mix of vector-based and transition-based strategies
Table 1
Summary of the input vector received by deep RL agents, showing the information provided to the agents and the length of each information encoded.