Randomly masking and predicting word tokens has been a successful approach in pre-training language models for a variety of downstream tasks. In this work, we observe that the same idea also applies naturally to sequential decision making, where many well-studied tasks like behavior cloning, offline RL, inverse dynamics, and waypoint conditioning correspond to different sequence maskings over a sequence of states, actions, and returns. We introduce the UniMASK framework, which provides a unified way to specify models which can be trained on many different sequential decision making tasks. We show that a single UniMASK model is often capable of carrying out many tasks with performance similar to or better than single-task models. Additionally, after fine-tuning, our UniMASK models consistently outperform comparable single-task models.

Simulated humanoids are an appealing research domain due to their physical capabilities. Nonetheless, they are also challenging to control, as a policy must drive an unstable, discontinuous, and high-dimensional physical system. One widely studied approach is to utilize motion capture (MoCap) data to teach the humanoid agent low-level skills (e.g., standing, walking, and running) that can then be re-used to synthesize high-level behaviors. However, even with MoCap data, controlling simulated humanoids remains very hard, as MoCap data offers only kinematic information. Finding physical control inputs to realize the demonstrated motions requires computationally intensive methods like reinforcement learning. Thus, despite the publicly available MoCap data, its utility has been limited to institutions with large-scale compute. In this work, we dramatically lower the barrier for productive research on this topic by training and releasing high-quality agents that can track over three hours of MoCap data for a simulated humanoid in the dmcontrol physics-based environment. We release MoCapAct (Motion Capture with Actions), a dataset of these expert agents and their rollouts, which contain proprioceptive observations and actions. We demonstrate the utility of MoCapAct by using it to train a single hierarchical policy capable of tracking the entire MoCap dataset within dmcontrol and show the learned low-level component can be re-used to efficiently learn downstream high-level tasks. Finally, we use MoCapAct to train an autoregressive GPT model and show that it can control a simulated humanoid to perform natural motion completion given a motion prompt.Videos of the results and links to the code and dataset are available at https://microsoft.github.io/MoCapAct.

Shunyu Yao, Karthik Narasimhan, Matthew Hausknecht. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

A hallmark of human intelligence is the ability to understand and communicate with language. Interactive Fiction games are fully text-based simulation environments where a player issues text commands to effect change in the environment and progress through the story. We argue that IF games are an excellent testbed for studying language-based autonomous agents. In particular, IF games combine challenges of combinatorial action spaces, language understanding, and commonsense reasoning. To facilitate rapid development of language-based agents, we introduce Jericho, a learning environment for man-made IF games and conduct a comprehensive study of text-agents across a rich set of games, highlighting directions in which agents can improve.

Text-based games present a unique challenge for autonomous agents to operate in natural language and handle enormous action spaces.In this paper, we propose the Contextual Action Language Model (CALM) to generate a compact set of action candidates at each game state.Our key insight is to train language models on human gameplay, where people demonstrate linguistic priors and a general game sense for promising actions conditioned on game history.We combine CALM with a reinforcement learning agent which re-ranks the generated action candidates to maximize ingame rewards.We evaluate our approach using the Jericho benchmark (Hausknecht et al., 2019a), on games unseen by CALM during training.Our method obtains a 69% relative improvement in average game score over the previous state-of-the-art model.Surprisingly, on half of these games, CALM is competitive with or better than other models that have access to ground truth admissible actions.* * Code and data are available at https://github. com/princeton-nlp/calm-textgame.Observation: You are in the living room.There is a doorway to the east, a wooden door with strange gothic lettering to the west, which appears to be nailed shut, a trophy case, and a large oriental rug in the center of the room.You are carrying: A brass lantern . . .

Transformers have increasingly outperformed gated RNNs in obtaining new state-of-the-art results on supervised tasks involving text sequences. Inspired by this trend, we study the question of how Transformer-based models can improve the performance of sequential decision-making agents. We present the Working Memory Graph (WMG), an agent that employs multi-head self-attention to reason over a dynamic set of vectors representing observed and recurrent state. We evaluate WMG in three environments featuring factored observation spaces: a Pathfinding environment that requires complex reasoning over past observations, BabyAI gridworld levels that involve variable goals, and Sokoban which emphasizes future planning. We find that the combination of WMG’s Transformer-based architecture with factored observation spaces leads to significant gains in learning efficiency compared to baseline architectures across all tasks. WMG demonstrates how Transformer-based models can dramatically boost sample efficiency in RL environments for which observations can be factored.

We study the problem of controllable generation of long-term sequential behaviors, where the goal is to calibrate to multiple behavior styles simultaneously. In contrast to the well-studied areas of controllable generation of images, text, and speech, there are two questions that pose significant challenges when generating long-term behaviors: how should we specify the factors of variation to control, and how can we ensure that the generated behavior faithfully demonstrates combinatorially many styles? We leverage programmatic labeling functions to specify controllable styles, and derive a formal notion of style-consistency as a learning objective, which can then be solved using conventional policy learning approaches. We evaluate our framework using demonstrations from professional basketball players and agents in the MuJoCo physics environment, and show that existing approaches that do not explicitly enforce style-consistency fail to generate diverse behaviors whereas our learned policies can be calibrated for up to $4^5 (1024)$ distinct style combinations.

The Arcade Learning Environment (ALE) is an evaluation platform that poses the challenge of building AI agents with general competency across dozens of Atari 2600 games. It supports a variety of different problem settings and it has been receiving increasing attention from the scientific community. In this paper we take a big picture look at how the ALE is being used by the research community. We focus on how diverse the evaluation methodologies in the ALE have become and we highlight some key concerns when evaluating agents in this platform. We use this discussion to present what we consider to be the best practices for future evaluations in the ALE. To further the progress in the field, we also introduce a new version of the ALE that supports multiple game modes and provides a form of stochasticity we call sticky actions.

The Arcade Learning Environment (ALE) is an evaluation platform that poses the challenge of building AI agents with general competency across dozens of Atari 2600 games. It supports a variety of different problem settings and it has been receiving increasing attention from the scientific community, leading to some high-profile success stories such as the much publicized Deep Q-Networks (DQN). In this article we take a big picture look at how the ALE is being used by the research community. We show how diverse the evaluation methodologies in the ALE have become with time, and highlight some key concerns when evaluating agents in the ALE. We use this discussion to present some methodological best practices and provide new benchmark results using these best practices. To further the progress in the field, we introduce a new version of the ALE that supports multiple game modes and provides a form of stochasticity we call sticky actions. We conclude this big picture look by revisiting challenges posed when the ALE was introduced, summarizing the state-of-the-art in various problems and highlighting problems that remain open.

Most recently proposed methods for Neural Program induction work under the assumption of having a large set of input/output (I/O) examples for learning any given input-output mapping. This paper aims to address the problem of data and computation efficiency of program induction by leveraging information from related tasks. Specifically, we propose two novel approaches for cross-task knowledge transfer to improve program induction in limited-data scenarios. In our first proposal, portfolio adaptation, a set of induction models is pretrained on a set of related tasks, and the best model is adapted towards the new task using transfer learning. In our second approach, meta program induction, a $k$-shot learning approach is used to make a model generalize to new tasks without additional training. To test the efficacy of our methods, we constructed a new benchmark of programs written in the Karel programming language. Using an extensive experimental evaluation on the Karel benchmark, we demonstrate that our proposals dramatically outperform the baseline induction method that does not use knowledge transfer. We also analyze the relative performance of the two approaches and study conditions in which they perform best. In particular, meta induction outperforms all existing approaches under extreme data sparsity (when a very small number of examples are available), i.e., fewer than ten. As the number of available I/O examples increase (i.e. a thousand or more), portfolio adapted program induction becomes the best approach. For intermediate data sizes, we demonstrate that the combined method of adapted meta program induction has the strongest performance.

This paper describes the learning and control capabilities of a biologically constrained bottom-up model of the mammalian cerebellum. Results are presented from six tasks: 1) eyelid conditioning; 2) pendulum balancing; 3) proportional-integral-derivative control; 4) robot balancing; 5) pattern recognition; and 6) MNIST handwritten digit recognition. These tasks span several paradigms of machine learning, including supervised learning, reinforcement learning, control, and pattern recognition. Results over these six domains indicate that the cerebellar simulation is capable of robustly identifying static input patterns even when randomized across the sensory apparatus. This capability allows the simulated cerebellum to perform several different supervised learning and control tasks. On the other hand, both reinforcement learning and temporal pattern recognition prove problematic due to the delayed nature of error signals and the simulator's inability to solve the credit assignment problem. These results are consistent with previous findings which hypothesize that in the human brain, the basal ganglia is responsible for reinforcement learning, while the cerebellum handles supervised learning.

Convolutional neural networks (CNNs) have been extensively applied for image recognition problems giving state-of-the-art results on recognition, detection, segmentation and retrieval. In this work we propose and evaluate several deep neural network architectures to combine image information across a video over longer time periods than previously attempted. We propose two methods capable of handling full length videos. The first method explores various convolutional temporal feature pooling architectures, examining the various design choices which need to be made when adapting a CNN for this task. The second proposed method explicitly models the video as an ordered sequence of frames. For this purpose we employ a recurrent neural network that uses Long Short-Term Memory (LSTM) cells which are connected to the output of the underlying CNN. Our best networks exhibit significant performance improvements over previously published results on the Sports 1 million dataset (73.1% vs. 60.9%) and the UCF-101 datasets with (88.6% vs. 88.0%) and without additional optical flow information (82.6% vs. 73.0%).

This paper addresses the challenge of learning to play many different video games with little domain-specific knowledge. Specifically, it introduces a neuroevolution approach to general Atari 2600 game playing. Four neuroevolution algorithms were paired with three different state representations and evaluated on a set of 61 Atari games. The neuroevolution agents represent different points along the spectrum of algorithmic sophistication - including weight evolution on topologically fixed neural networks (conventional neuroevolution), covariance matrix adaptation evolution strategy (CMA-ES), neuroevolution of augmenting topologies (NEAT), and indirect network encoding (HyperNEAT). State representations include an object representation of the game screen, the raw pixels of the game screen, and seeded noise (a comparative baseline). Results indicate that direct-encoding methods work best on compact state representations while indirect-encoding methods (i.e., HyperNEAT) allow scaling to higher dimensional representations (i.e., the raw game screen). Previous approaches based on temporal-difference (TD) learning had trouble dealing with the large state spaces and sparse reward gradients often found in Atari games. Neuroevolution ameliorates these problems and evolved policies achieve state-of-the-art results, even surpassing human high scores on three games. These results suggest that neuroevolution is a promising approach to general video game playing (GVGP).

This paper considers the challenge of enabling agents to learn with as little domain-specific knowledge as possible. The main contribution is HyperNEAT-GGP, a HyperNEAT-based General Game Playing approach to Atari games. By leveraging the geometric regularities present in the Atari game screen, HyperNEAT effectively evolves policies for playing two different Atari games, Asterix and Freeway. Results show that HyperNEAT-GGP outperforms existing benchmarks on these games. HyperNEAT-GGP represents a step towards the ambitious goal of creating an agent capable of learning and seamlessly transitioning between many different tasks.

Advances in autonomous vehicles and intelligent transportation systems indicate a rapidly approaching future in which intelligent vehicles will automatically handle the process of driving. However, increasing the efficiency of today's transportation infrastructure will require intelligent traffic control mechanisms that work hand in hand with intelligent vehicles. To this end, Dresner and Stone proposed a new intersection control mechanism called Autonomous Intersection Management (AIM) and showed in simulation that by studying the problem from a multiagent perspective, intersection control can be made more efficient than existing control mechanisms such as traffic signals and stop signs. We extend their study beyond the case of an individual intersection and examine the unique implications and abilities afforded by using AIM-based agents to control a network of interconnected intersections. We examine different navigation policies by which autonomous vehicles can dynamically alter their planned paths, observe an instance of Braess' paradox, and explore the new possibility of dynamically reversing the flow of traffic along lanes in response to minute-by-minute traffic conditions. Studying this multiagent system in simulation, we quantify the substantial improvements in efficiency imparted by these agent-based traffic control methods.

It is generally agreed that there is no valid reason to use a long connection chain for remote login such as SSH connection. Most of the stepping-stone detection algorithms installed on a host were designed to protect the victim of a third party downstream from where the algorithm is running. It is much more important for a host to protect itself from being a victim. This project uses an approximated round-trip time to distinguish a long connection chain from a short one. Several measures were studied to distinguish long chains from short ones. An estimated roundtrip time was defined to measure the chain length. Preliminary result suggests shows that the proposed algorithm can distinguish long connection chains from short ones with relatively low false rate.