Atari-GPT: Investigating the Capabilities of Multimodal Large Language Models as Low-Level Policies for Atari Games

Recent advancements in large language models (LLMs) have expanded their capabilities beyond traditional text-based tasks to multimodal domains, integrating visual, auditory, and textual data. While multimodal LLMs have been extensively explored for high-level planning in domains like robotics and games, their potential as low-level controllers remains largely untapped. This paper explores the application of multimodal LLMs as low-level controllers in the domain of Atari video games, introducing Atari game performance as a new benchmark for evaluating the ability of multimodal LLMs to perform low-level control tasks. Unlike traditional reinforcement learning (RL) and imitation learning (IL) methods that require extensive computational resources as well as reward function specification, these LLMs utilize pre-existing multimodal knowledge to directly engage with game environments. Our study assesses multiple multimodal LLMs performance against traditional RL agents, human players, and random agents, focusing on their ability to understand and interact with complex visual scenes and formulate strategic responses. Additionally, we examine the impact of In-Context Learning (ICL) by incorporating human-demonstrated game-play trajectories to enhance the models contextual understanding. Through this investigation, we aim to determine the extent to which multimodal LLMs can leverage their extensive training to effectively function as low-level controllers, thereby redefining potential applications in dynamic and visually complex environments. Additional results and videos are available at our project webpage: https://sites.google.com/view/atari-gpt/.

StarCraftImage: A Dataset For Prototyping Spatial Reasoning Methods For Multi-Agent Environments

Spatial reasoning tasks in multi-agent environments such as event prediction, agent type identification, or missing data imputation are important for multiple applications (e.g., autonomous surveillance over sensor networks and subtasks for reinforcement learning (RL)). StarCraft II game replays encode intelligent (and adversarial) multi-agent behavior and could provide a testbed for these tasks; however, extracting simple and standardized representations for prototyping these tasks is laborious and hinders reproducibility. In contrast, MNIST and CIFAR10, despite their extreme simplicity, have enabled rapid prototyping and reproducibility of ML methods. Following the simplicity of these datasets, we construct a benchmark spatial reasoning dataset based on StarCraft II replays that exhibit complex multi-agent behaviors, while still being as easy to use as MNIST and CIFAR10. Specifically, we carefully summarize a window of 255 consecutive game states to create 3.6 million summary images from 60,000 replays, including all relevant metadata such as game outcome and player races. We develop three formats of decreasing complexity: Hyperspectral images that include one channel for every unit type (similar to multispectral geospatial images), RGB images that mimic CIFAR10, and grayscale images that mimic MNIST. We show how this dataset can be used for prototyping spatial reasoning methods. All datasets, code for extraction, and code for dataset loading can be found at https://starcraftdata.davidinouye.com

DisasterResponseGPT: Large Language Models for Accelerated Plan of Action Development in Disaster Response Scenarios

The development of plans of action in disaster response scenarios is a time-consuming process. Large Language Models (LLMs) offer a powerful solution to expedite this process through in-context learning. This study presents DisasterResponseGPT, an algorithm that leverages LLMs to generate valid plans of action quickly by incorporating disaster response and planning guidelines in the initial prompt. In DisasterResponseGPT, users input the scenario description and receive a plan of action as output. The proposed method generates multiple plans within seconds, which can be further refined following the user's feedback. Preliminary results indicate that the plans of action developed by DisasterResponseGPT are comparable to human-generated ones while offering greater ease of modification in real-time. This approach has the potential to revolutionize disaster response operations by enabling rapid updates and adjustments during the plan's execution.

Learning Flight Control Systems from Human Demonstrations and Real-Time Uncertainty-Informed Interventions

This paper describes a methodology for learning flight control systems from human demonstrations and interventions while considering the estimated uncertainty in the learned models. The proposed approach uses human demonstrations to train an initial model via imitation learning and then iteratively, improve its performance by using real-time human interventions. The aim of the interventions is to correct undesired behaviors and adapt the model to changes in the task dynamics. The learned model uncertainty is estimated in real-time via Monte Carlo Dropout and the human supervisor is cued for intervention via an audiovisual signal when this uncertainty exceeds a predefined threshold. This proposed approach is validated in an autonomous quadrotor landing task on both fixed and moving platforms. It is shown that with this algorithm, a human can rapidly teach a flight task to an unmanned aerial vehicle via demonstrating expert trajectories and then adapt the learned model by intervening when the learned controller performs any undesired maneuver, the task changes, and/or the model uncertainty exceeds a threshold

Gaze-Informed Multi-Objective Imitation Learning from Human Demonstrations

In the field of human-robot interaction, teaching learning agents from human demonstrations via supervised learning has been widely studied and successfully applied to multiple domains such as self-driving cars and robot manipulation. However, the majority of the work on learning from human demonstrations utilizes only behavioral information from the demonstrator, i.e. what actions were taken, and ignores other useful information. In particular, eye gaze information can give valuable insight towards where the demonstrator is allocating their visual attention, and leveraging such information has the potential to improve agent performance. Previous approaches have only studied the utilization of attention in simple, synchronous environments, limiting their applicability to real-world domains. This work proposes a novel imitation learning architecture to learn concurrently from human action demonstration and eye tracking data to solve tasks where human gaze information provides important context. The proposed method is applied to a visual navigation task, in which an unmanned quadrotor is trained to search for and navigate to a target vehicle in a real-world, photorealistic simulated environment. When compared to a baseline imitation learning architecture, results show that the proposed gaze augmented imitation learning model is able to learn policies that achieve significantly higher task completion rates, with more efficient paths, while simultaneously learning to predict human visual attention. This research aims to highlight the importance of multimodal learning of visual attention information from additional human input modalities and encourages the community to adopt them when training agents from human demonstrations to perform visuomotor tasks.

PODNet: A Neural Network for Discovery of Plannable Options

Learning from demonstration has been widely studied in machine learning but becomes challenging when the demonstrated trajectories are unstructured and follow different objectives. This short-paper proposes PODNet, Plannable Option Discovery Network, addressing how to segment an unstructured set of demonstrated trajectories for option discovery. This enables learning from demonstration to perform multiple tasks and plan high-level trajectories based on the discovered option labels. PODNet combines a custom categorical variational autoencoder, a recurrent option inference network, option-conditioned policy network, and option dynamics model in an end-to-end learning architecture. Due to the concurrently trained option-conditioned policy network and option dynamics model, the proposed architecture has implications in multi-task and hierarchical learning, explainable and interpretable artificial intelligence, and applications where the agent is required to learn only from observations.

Integrating Behavior Cloning and Reinforcement Learning for Improved Performance in Sparse Reward Environments

This paper investigates how to efficiently transition and update policies, trained initially with demonstrations, using off-policy actor-critic reinforcement learning. It is well-known that techniques based on Learning from Demonstrations, for example behavior cloning, can lead to proficient policies given limited data. However, it is currently unclear how to efficiently update that policy using reinforcement learning as these approaches are inherently optimizing different objective functions. Previous works have used loss functions which combine behavioral cloning losses with reinforcement learning losses to enable this update, however, the components of these loss functions are often set anecdotally, and their individual contributions are not well understood. In this work we propose the Cycle-of-Learning (CoL) framework that uses an actor-critic architecture with a loss function that combines behavior cloning and 1-step Q-learning losses with an off-policy pre-training step from human demonstrations. This enables transition from behavior cloning to reinforcement learning without performance degradation and improves reinforcement learning in terms of overall performance and training time. Additionally, we carefully study the composition of these combined losses and their impact on overall policy learning. We show that our approach outperforms state-of-the-art techniques for combining behavior cloning and reinforcement learning for both dense and sparse reward scenarios. Our results also suggest that directly including the behavior cloning loss on demonstration data helps to ensure stable learning and ground future policy updates.

On Memory Mechanism in Multi-Agent Reinforcement Learning

Multi-agent reinforcement learning (MARL) extends (single-agent) reinforcement learning (RL) by introducing additional agents and (potentially) partial observability of the environment. Consequently, algorithms for solving MARL problems incorporate various extensions beyond traditional RL methods, such as a learned communication protocol between cooperative agents that enables exchange of private information or adaptive modeling of opponents in competitive settings. One popular algorithmic construct is a memory mechanism such that an agent's decisions can depend not only upon the current state but also upon the history of observed states and actions. In this paper, we study how a memory mechanism can be useful in environments with different properties, such as observability, internality and presence of a communication channel. Using both prior work and new experiments, we show that a memory mechanism is helpful when learning agents need to model other agents and/or when communication is constrained in some way; however we must to be cautious of agents achieving effective memoryfulness through other means.

On the use of Deep Autoencoders for Efficient Embedded Reinforcement Learning

In autonomous embedded systems, it is often vital to reduce the amount of actions taken in the real world and energy required to learn a policy. Training reinforcement learning agents from high dimensional image representations can be very expensive and time consuming. Autoencoders are deep neural network used to compress high dimensional data such as pixelated images into small latent representations. This compression model is vital to efficiently learn policies, especially when learning on embedded systems. We have implemented this model on the NVIDIA Jetson TX2 embedded GPU, and evaluated the power consumption, throughput, and energy consumption of the autoencoders for various CPU/GPU core combinations, frequencies, and model parameters. Additionally, we have shown the reconstructions generated by the autoencoder to analyze the quality of the generated compressed representation and also the performance of the reinforcement learning agent. Finally, we have presented an assessment of the viability of training these models on embedded systems and their usefulness in developing autonomous policies. Using autoencoders, we were able to achieve 4-5 $\times$ improved performance compared to a baseline RL agent with a convolutional feature extractor, while using less than 2W of power.

Efficiently Combining Human Demonstrations and Interventions for Safe Training of Autonomous Systems in Real-Time

This paper investigates how to utilize different forms of human interaction to safely train autonomous systems in real-time by learning from both human demonstrations and interventions. We implement two components of the Cycle-of-Learning for Autonomous Systems, which is our framework for combining multiple modalities of human interaction. The current effort employs human demonstrations to teach a desired behavior via imitation learning, then leverages intervention data to correct for undesired behaviors produced by the imitation learner to teach novel tasks to an autonomous agent safely, after only minutes of training. We demonstrate this method in an autonomous perching task using a quadrotor with continuous roll, pitch, yaw, and throttle commands and imagery captured from a downward-facing camera in a high-fidelity simulated environment. Our method improves task completion performance for the same amount of human interaction when compared to learning from demonstrations alone, while also requiring on average 32% less data to achieve that performance. This provides evidence that combining multiple modes of human interaction can increase both the training speed and overall performance of policies for autonomous systems.