Games are challenging for Reinforcement Learning~(RL) agents due to their reward-sparsity, as rewards are only obtainable after long sequences of deliberate actions. Intrinsic Motivation~(IM) methods -- which introduce exploration rewards -- are an effective solution to reward-sparsity. However, IM also causes an issue known as `reward hacking' where the agent optimizes for the new reward at the expense of properly playing the game. The larger problem is that reward hacking itself is largely unknown; there is no answer to whether, and to what extent, IM rewards change the behavior of RL agents. This study takes a first step by empirically evaluating the impact on behavior of three IM techniques on the MiniGrid game-like environment. We compare these IM models with Generalized Reward Matching~(GRM), a method that can be used with any intrinsic reward function to guarantee optimality. Our results suggest that IM causes noticeable change by increasing the initial rewards, but also altering the way the agent plays; and that GRM mitigated reward hacking in some scenarios.

This paper investigates two fundamental problems that arise when utilizing Intrinsic Motivation (IM) for reinforcement learning in Reward-Free Pre-Training (RFPT) tasks and Exploration with Intrinsic Motivation (EIM) tasks: 1) how to design an effective intrinsic objective in RFPT tasks, and 2) how to reduce the bias introduced by the intrinsic objective in EIM tasks. Existing IM methods suffer from static skills, limited state coverage, sample inefficiency in RFPT tasks, and suboptimality in EIM tasks. To tackle these problems, we propose \emph{Constrained Intrinsic Motivation (CIM)} for RFPT and EIM tasks, respectively: 1) CIM for RFPT maximizes the lower bound of the conditional state entropy subject to an alignment constraint on the state encoder network for efficient dynamic and diverse skill discovery and state coverage maximization; 2) CIM for EIM leverages constrained policy optimization to adaptively adjust the coefficient of the intrinsic objective to mitigate the distraction from the intrinsic objective. In various MuJoCo robotics environments, we empirically show that CIM for RFPT greatly surpasses fifteen IM methods for unsupervised skill discovery in terms of skill diversity, state coverage, and fine-tuning performance. Additionally, we showcase the effectiveness of CIM for EIM in redeeming intrinsic rewards when task rewards are exposed from the beginning. Our code is available at https://github.com/x-zheng16/CIM.