Nicholas Roy

Nicholas A. Roy, Junkyung Kim, Neil Rabinowitz

Explaining the behavior of AI systems is an important problem that, in practice, is generally avoided. While the XAI community has been developing an abundance of techniques, most incur a set of costs that the wider deep learning community has been unwilling to pay in most situations. We take a pragmatic view of the issue, and define a set of desiderata that capture both the ambitions of XAI and the practical constraints of deep learning. We describe an effective way to satisfy all the desiderata: train the AI system to build a causal model of itself. We develop an instance of this solution for Deep RL agents: Causal Self-Talk. CST operates by training the agent to communicate with itself across time. We implement this method in a simulated 3D environment, and show how it enables agents to generate faithful and semantically-meaningful explanations of their own behavior. Beyond explanations, we also demonstrate that these learned models provide new ways of building semantic control interfaces to AI systems.

Ethan Pronovost, Meghana Reddy Ganesina, Noureldin Hendy et al.

Automated creation of synthetic traffic scenarios is a key part of validating the safety of autonomous vehicles (AVs). In this paper, we propose Scenario Diffusion, a novel diffusion-based architecture for generating traffic scenarios that enables controllable scenario generation. We combine latent diffusion, object detection and trajectory regression to generate distributions of synthetic agent poses, orientations and trajectories simultaneously. To provide additional control over the generated scenario, this distribution is conditioned on a map and sets of tokens describing the desired scenario. We show that our approach has sufficient expressive capacity to model diverse traffic patterns and generalizes to different geographical regions.

Martina Stadler Kurtz, Samuel Prentice, Yasmin Veys et al.

We would like to enable a collaborative multiagent team to navigate at long length scales and under uncertainty in real-world environments. In practice, planning complexity scales with the number of agents in the team, with the length scale of the environment, and with environmental uncertainty. Enabling tractable planning requires developing abstract models that can represent complex, high-quality plans. However, such models often abstract away information needed to generate directly-executable plans for real-world agents in real-world environments, as planning in such detail, especially in the presence of real-world uncertainty, would be computationally intractable. In this paper, we describe the deployment of a planning system that used a hierarchy of planners to execute collaborative multiagent navigation tasks in real-world, unknown environments. By developing a planning system that was robust to failures at every level of the planning hierarchy, we enabled the team to complete collaborative navigation tasks, even in the presence of imperfect planning abstractions and real-world uncertainty. We deployed our approach on a Clearpath Husky-Jackal team navigating in a structured outdoor environment, and demonstrated that the system enabled the agents to successfully execute collaborative plans.

Aaron Ray, Jacob Arkin, Harel Biggie et al.

In order to provide a robot with the ability to understand and react to a user's natural language inputs, the natural language must be connected to the robot's underlying representations of the world. Recently, large language models (LLMs) and 3D scene graphs (3DSGs) have become a popular choice for grounding natural language and representing the world. In this work, we address the challenge of using LLMs with 3DSGs to ground natural language. Existing methods encode the scene graph as serialized text within the LLM's context window, but this encoding does not scale to large or rich 3DSGs. Instead, we propose to use a form of Retrieval Augmented Generation to select a subset of the 3DSG relevant to the task. We encode a 3DSG in a graph database and provide a query language interface (Cypher) as a tool to the LLM with which it can retrieve relevant data for language grounding. We evaluate our approach on instruction following and scene question-answering tasks and compare against baseline context window and code generation methods. Our results show that using Cypher as an interface to 3D scene graphs scales significantly better to large, rich graphs on both local and cloud-based models. This leads to large performance improvements in grounded language tasks while also substantially reducing the token count of the scene graph content. A video supplement is available at https://www.youtube.com/watch?v=zY_YI9giZSA.

Ethan Pronovost, Neha Boloor, Peter Schleede et al.

Processing spatial data is a key component in many learning tasks for autonomous driving such as motion forecasting, multi-agent simulation, and planning. Prior works have demonstrated the value in using SE(2) invariant network architectures that consider only the relative poses between objects (e.g. other agents, scene features such as traffic lanes). However, these methods compute the relative poses for all pairs of objects explicitly, requiring quadratic memory. In this work, we propose a mechanism for SE(2) invariant scaled dot-product attention that requires linear memory relative to the number of objects in the scene. Our SE(2) invariant transformer architecture enjoys the same scaling properties that have benefited large language models in recent years. We demonstrate experimentally that our approach is practical to implement and improves performance compared to comparable non-invariant architectures.

Emily Liu, Michael Noseworthy, Nicholas Roy

The scarcity of labeled action data poses a considerable challenge for developing machine learning algorithms for robotic object manipulation. It is expensive and often infeasible for a robot to interact with many objects. Conversely, visual data of objects, without interaction, is abundantly available and can be leveraged for pretraining and feature extraction. However, current methods that rely on image data for pretraining do not easily adapt to task-specific predictions, since the learned features are not guaranteed to be relevant. This paper introduces the Semi-Supervised Neural Process (SSNP): an adaptive reward-prediction model designed for scenarios in which only a small subset of objects have labeled interaction data. In addition to predicting reward labels, the latent-space of the SSNP is jointly trained with an autoencoding objective using passive data from a much larger set of objects. Jointly training with both types of data allows the model to focus more effectively on generalizable features and minimizes the need for extensive retraining, thereby reducing computational demands. The efficacy of SSNP is demonstrated through a door-opening task, leading to better performance than other semi-supervised methods, and only using a fraction of the data compared to other adaptive models.

Andrew K. Lampinen, Nicholas A. Roy, Ishita Dasgupta et al.

Inferring the abstract relational and causal structure of the world is a major challenge for reinforcement-learning (RL) agents. For humans, language--particularly in the form of explanations--plays a considerable role in overcoming this challenge. Here, we show that language can play a similar role for deep RL agents in complex environments. While agents typically struggle to acquire relational and causal knowledge, augmenting their experience by training them to predict language descriptions and explanations can overcome these limitations. We show that language can help agents learn challenging relational tasks, and examine which aspects of language contribute to its benefits. We then show that explanations can help agents to infer not only relational but also causal structure. Language can shape the way that agents to generalize out-of-distribution from ambiguous, causally-confounded training, and explanations even allow agents to learn to perform experimental interventions to identify causal relationships. Our results suggest that language description and explanation may be powerful tools for improving agent learning and generalization.

Caris Moses, Michael Noseworthy, Leslie Pack Kaelbling et al.

Exploration in novel settings can be challenging without prior experience in similar domains. However, humans are able to build on prior experience quickly and efficiently. Children exhibit this behavior when playing with toys. For example, given a toy with a yellow and blue door, a child will explore with no clear objective, but once they have discovered how to open the yellow door, they will most likely be able to open the blue door much faster. Adults also exhibit this behavior when entering new spaces such as kitchens. We develop a method, Contextual Prior Prediction, which provides a means of transferring knowledge between interactions in similar domains through vision. We develop agents that exhibit exploratory behavior with increasing efficiency, by learning visual features that are shared across environments, and how they correlate to actions. Our problem is formulated as a Contextual Multi-Armed Bandit where the contexts are images, and the robot has access to a parameterized action space. Given a novel object, the objective is to maximize reward with few interactions. A domain which strongly exhibits correlations between visual features and motion is kinemetically constrained mechanisms. We evaluate our method on simulated prismatic and revolute joints.

Gregory J. Stein, Christopher Bradley, Victoria Preston et al.

Topological strategies for navigation meaningfully reduce the space of possible actions available to a robot, allowing use of heuristic priors or learning to enable computationally efficient, intelligent planning. The challenges in estimating structure with monocular SLAM in low texture or highly cluttered environments have precluded its use for topological planning in the past. We propose a robust sparse map representation that can be built with monocular vision and overcomes these shortcomings. Using a learned sensor, we estimate high-level structure of an environment from streaming images by detecting sparse vertices (e.g., boundaries of walls) and reasoning about the structure between them. We also estimate the known free space in our map, a necessary feature for planning through previously unknown environments. We show that our mapping technique can be used on real data and is sufficient for planning and exploration in simulated multi-agent search and learned subgoal planning applications.

Rohan Paul, Andrei Barbu, Sue Felshin et al.

A robot's ability to understand or ground natural language instructions is fundamentally tied to its knowledge about the surrounding world. We present an approach to grounding natural language utterances in the context of factual information gathered through natural-language interactions and past visual observations. A probabilistic model estimates, from a natural language utterance, the objects,relations, and actions that the utterance refers to, the objectives for future robotic actions it implies, and generates a plan to execute those actions while updating a state representation to include newly acquired knowledge from the visual-linguistic context. Grounding a command necessitates a representation for past observations and interactions; however, maintaining the full context consisting of all possible observed objects, attributes, spatial relations, actions, etc., over time is intractable. Instead, our model, Temporal Grounding Graphs, maintains a learned state representation for a belief over factual groundings, those derived from natural-language interactions, and lazily infers new groundings from visual observations using the context implied by the utterance. This work significantly expands the range of language that a robot can understand by incorporating factual knowledge and observations of its workspace in its inference about the meaning and grounding of natural-language utterances.

Genevieve Flaspohler, Nicholas Roy, Yogesh Girdhar

The gap between our ability to collect interesting data and our ability to analyze these data is growing at an unprecedented rate. Recent algorithmic attempts to fill this gap have employed unsupervised tools to discover structure in data. Some of the most successful approaches have used probabilistic models to uncover latent thematic structure in discrete data. Despite the success of these models on textual data, they have not generalized as well to image data, in part because of the spatial and temporal structure that may exist in an image stream. We introduce a novel unsupervised machine learning framework that incorporates the ability of convolutional autoencoders to discover features from images that directly encode spatial information, within a Bayesian nonparametric topic model that discovers meaningful latent patterns within discrete data. By using this hybrid framework, we overcome the fundamental dependency of traditional topic models on rigidly hand-coded data representations, while simultaneously encoding spatial dependency in our topics without adding model complexity. We apply this model to the motivating application of high-level scene understanding and mission summarization for exploratory marine robots. Our experiments on a seafloor dataset collected by a marine robot show that the proposed hybrid framework outperforms current state-of-the-art approaches on the task of unsupervised seafloor terrain characterization.

Sachithra Hemachandra, Felix Duvallet, Thomas M. Howard et al.

Natural language offers an intuitive and flexible means for humans to communicate with the robots that we will increasingly work alongside in our homes and workplaces. Recent advancements have given rise to robots that are able to interpret natural language manipulation and navigation commands, but these methods require a prior map of the robot's environment. In this paper, we propose a novel learning framework that enables robots to successfully follow natural language route directions without any previous knowledge of the environment. The algorithm utilizes spatial and semantic information that the human conveys through the command to learn a distribution over the metric and semantic properties of spatially extended environments. Our method uses this distribution in place of the latent world model and interprets the natural language instruction as a distribution over the intended behavior. A novel belief space planner reasons directly over the map and behavior distributions to solve for a policy using imitation learning. We evaluate our framework on a voice-commandable wheelchair. The results demonstrate that by learning and performing inference over a latent environment model, the algorithm is able to successfully follow natural language route directions within novel, extended environments.

Joshua Joseph, Javier Velez, Nicholas Roy

Batch Reinforcement Learning (RL) algorithms attempt to choose a policy from a designer-provided class of policies given a fixed set of training data. Choosing the policy which maximizes an estimate of return often leads to over-fitting when only limited data is available, due to the size of the policy class in relation to the amount of data available. In this work, we focus on learning policy classes that are appropriately sized to the amount of data available. We accomplish this by using the principle of Structural Risk Minimization, from Statistical Learning Theory, which uses Rademacher complexity to identify a policy class that maximizes a bound on the return of the best policy in the chosen policy class, given the available data. Unlike similar batch RL approaches, our bound on return requires only extremely weak assumptions on the true system.

Alborz Geramifard, Thomas J. Walsh, Nicholas Roy et al.

Matching pursuit (MP) methods are a promising class of feature construction algorithms for value function approximation. Yet existing MP methods require creating a pool of potential features, mandating expert knowledge or enumeration of a large feature pool, both of which hinder scalability. This paper introduces batch incremental feature dependency discovery (Batch-iFDD) as an MP method that inherits a provable convergence property. Additionally, Batch-iFDD does not require a large pool of features, leading to lower computational complexity. Empirical policy evaluation results across three domains with up to one million states highlight the scalability of Batch-iFDD over the previous state of the art MP algorithm.