Reinforcement Learning (RL) has demonstrated remarkable success in solving sequential decision-making problems. However, in real-world scenarios, RL agents often struggle to generalize when faced with unseen actions that were not encountered during training. Some previous works on zero-shot action generalization rely on large datasets of action observations to capture the behaviors of new actions, making them impractical for real-world applications. In this paper, we introduce a novel zero-shot framework, Action Generalization from Limited Observations (AGLO). Our framework has two main components: an action representation learning module and a policy learning module. The action representation learning module extracts discriminative embeddings of actions from limited observations, while the policy learning module leverages the learned action representations, along with augmented synthetic action representations, to learn a policy capable of handling tasks with unseen actions. The experimental results demonstrate that our framework significantly outperforms state-of-the-art methods for zero-shot action generalization across multiple benchmark tasks, showcasing its effectiveness in generalizing to new actions with minimal action observations.

Simply showing that there is information about actions, actors (and their conjunction) in any part of the brain does not mean they have tackled a "(three-fold) challenge". One would find (presumably) very similar response properties in any part of the mirror neuron system. Furthermore, if one analysed retinal cells, one would also find this information. Perhaps you could highlight the fact that you have found information or invariance properties at the level of the single neuron - that could not be found at low levels in the visual hierarchy - to make your point more clearly?

A fundamental trait of intelligence is the ability to achieve goals in the face of novel circumstances, such as making decisions from new action choices. However, standard reinforcement learning assumes a fixed set of actions and requires expensive retraining when given a new action set. To make learning agents more adaptable, we introduce the problem of zero-shot generalization to new actions. We propose a two-stage framework where the agent first infers action representations from action information acquired separately from the task. A policy flexible to varying action sets is then trained with generalization objectives. We benchmark generalization on sequential tasks, such as selecting from an unseen tool-set to solve physical reasoning puzzles and stacking towers with novel 3D shapes. Videos and code are available at https://sites.google.com/view/action-generalization

It has been an open challenge for self-interested agents to make optimal sequential decisions in complex multiagent systems, where agents might achieve higher utility via collaboration. The Microsoft Malmo Collaborative AI Challenge (MCAC), which is designed to encourage research relating to various problems in Collaborative AI, takes the form of a Minecraft mini-game where players might work together to catch a pig or deviate from cooperation, for pursuing high scores to win the challenge. Various characteristics, such as complex interactions among agents, uncertainties, sequential decision making and limited learning trials all make it extremely challenging to find effective strategies. We present HogRider---the champion agent of MCAC in 2017 out of 81 teams from 26 countries. One key innovation of HogRider is a generalized agent type hypothesis framework to identify the behavior model of the other agents, which is demonstrated to be robust to observation uncertainty. On top of that, a second key innovation is a novel Q-learning approach to learn effective policies against each type of the collaborating agents. Various ideas are proposed to adapt traditional Q-learning to handle complexities in the challenge, including state-action abstraction to reduce problem scale, a warm start approach using human reasoning for addressing limited learning trials, and an active greedy strategy to balance exploitation-exploration. Challenge results show that HogRider outperforms all the other teams by a significant edge, in terms of both optimality and stability.

Potential games and decentralised partially observable MDPs (Dec-POMDPs) are two commonly used models of multi-agent interaction, for static optimisation and sequential decisionmaking settings, respectively. In this paper we introduce filtered fictitious play for solving repeated potential games in which each player's observations of others' actions are perturbed by random noise, and use this algorithm to construct an online learning method for solving Dec-POMDPs. Specifically, we prove that noise in observations prevents standard fictitious play from converging to Nash equilibrium in potential games, which also makes fictitious play impractical for solving Dec-POMDPs. To combat this, we derive filtered fictitious play, and provide conditions under which it converges to a Nash equilibrium in potential games with noisy observations. We then use filtered fictitious play to construct a solver for Dec-POMDPs, and demonstrate our new algorithm's performance in a box pushing problem. Our results show that we consistently outperform the state-of-the-art Dec-POMDP solver by an average of 100% across the range of noise in the observation function.