Research

Operator learning is a new frontier in computational learning-based approaches for autonomy that enables us to work directly with function-to-function mappings. We present Basis-to-Basis (B2B) operator learning, a novel approach for learning operators on Hilbert spaces of functions based on the foundational ideas of function encoders.

-Autonomous Systems Group

In traditional approaches to terrain-relative navigation (TRN), terrain measurements are compared to an elevation map carried onboard. These methods are computationally expensive, and it is impractical to store high-quality maps of large swaths of terrain. In this work, we generate position measurements using a novel method inspired by navigation in animals. We incorporate the approach into an inertial navigation scheme using a novel measurement rejection strategy and online covariance computation. Our results show that the filter provides accurate state information over the course of a long trajectory.

- Nonlinear Estimation and Autonomy Research Group

One of the fundamental challenges associated with reinforcement learning (RL) in robotics is that collecting sufficient data can be both time-consuming and expensive. In this paper, we formalize a concept of time reversal symmetry in Markov decision processes (MDPs) which builds upon the established structure of dynamically reversible Markov chains (DRMCs) and time reversibility in classical physics. Using this new construct, we investigate the utility of this concept in reducing the sample complexity of robot learning. We observe that utilizing the structure of time reversal in an MDP allows every environment transition experienced by an agent to be transformed into a feasible reverse-time transition, effectively doubling the number of experiences in the environment. To test the usefulness of this newly synthesized data, we introduce a novel approach called time symmetric data augmentation (TSDA) and investigate its application in both proprioceptive and pixel-based state within the realm of off-policy, model-free RL. Empirical evaluations showcase how these synthetic transitions can enhance the sample efficiency of RL agents in time reversible scenarios without friction or contact. We also test this method in less idealized environments, where TSDA can significantly degrade sample efficiency and policy performance, but can also improve sample efficiency under the right conditions. Ultimately we conclude that time symmetry shows promise in robot learning, but only if the environment and reward structure are of an appropriate form.

- Control and Learning for Autonomous Robotics Group

Researchers at the Center for Autonomy coauthored a NASA report on the vision for autonomy verification and validation. The report brings attention to verification challenges facing the design and deployment of autonomous aircraft in the emerging ecosystem of advanced air mobility. In addition, it proposes a future trajectory for solving these verification challenges as a multidisciplinary research community and conjectures on the enabled technologies that overcoming those challenges might give rise to. A particularly interesting development involves new algorithms and techniques that use learning to train on data posing unique challenges and constitute promising research trajectories within verifying and validating autonomy. Autonomous systems are also increasingly deployed alongside human agents, raising safety, assurance, and regulatory challenges. The report is joint work with industry (Boeing, Collins, and GE), government (NASA), and academia (UT Austin, MIT, and the University of Michigan), thereby developing a comprehensive understanding of these issues.

- Autonomous Systems Group

The project addresses the challenge of "safe and efficient growth in global operations, with the expected outcome of an algorithmic foundation that will help realize increasingly autonomous and collaborative air traffic management for all classes of airspace and vehicles and support scalable and efficient operations that rapidly adapt to meet changing demands and to respond to system disruptions.

- Autonomous Systems Group

In this article, we propose a path planning method- ology for a mobile robot navigating through an obstacle-filled environment to generate a reference path that is traceable with moderate sensing efforts. The desired reference path is characterized as the shortest path in an obstacle-filled Gaussian belief manifold equipped with a novel information-geometric distance function. The distance function we introduce is shown to be an asymmetric quasi-pseudometric and can be interpreted as the minimum information gain required to steer the Gaussian belief. An RRT*-based numerical solution algorithm is presented to solve the formulated shortest-path problem. To gain insight into the asymptotic optimality of the proposed algorithm, we show that the considered path length function is continuous with respect to the topology of total variation. Simulation results demonstrate that the proposed method is effective in various robot navigation scenarios to reduce sensing costs, such as the required frequency of sensor measurements and the number of sensors that must be operated simultaneously.

- Networked Control Systems Lab

In any competition, a key to a human team’s success is to be unpredictable. If a football team were to follow the same strategy to score a touchdown in every game, their competitors would exploit this information for their own advantage in the future. Can autonomous systems, like humans, act unpredictably to compete with their adversaries? Is there a limit on the unpredictability of their goal-directed behavior? Combining ideas from stochastic control and information theory, we establish the theoretical limits and computational requirements for unpredictable planning in adversarial environments. We also utilize these insights to develop efficient algorithms that enable autonomous systems to securely carry out tasks, e.g., surveillance, in the presence of adversaries.

- Autonomous Systems Group

3D printing of medium- to large-scale objects is often limited by the size of the 3D printer - the bigger the desired object, the bigger the 3D printer has to be. To address these limitations, we develop a 3D printing hexacopter that uses fused deposition modeling to deposit polymer materials during flight. Recently, we have demonstrated the feasibility of this approach by successfully printing contours of polylactic acid on a flat surface. Future research focuses on developing advanced control algorithms to improve the 3D printing accuracy.

- Autonomous Systems Group

A recent paper by members of the DCIST alliance develops a perception-action-communication loop framework using Vision-based Graph Aggregation and Inference (VGAI). This multi-agent decentralized learning-to-control framework maps raw visual observations to agent actions, aided by local communication among neighboring agents. The framework is implemented by a cascade of a convolutional and a graph neural network (CNN / GNN), addressing agent-level visual perception and feature learning, as well as swarm-level communication, local information aggregation and agent action inference, respectively. By jointly training the CNN and GNN, image features and communication messages are learned in conjunction to better address the specific task. The researchers use imitation learning to train the VGAI controller in an offline phase, relying on a centralized expert controller. This results in a learned VGAI controller that can be deployed in a distributed manner for online execution. Additionally, the controller exhibits good scaling properties, with training in smaller teams and application in larger teams. Through a multiagent flocking application, the researchers demonstrate that VGAI yields performance comparable to or better than other decentralized controllers, using only the visual input modality (even with visibility degradation) and without accessing precise location or motion state information.

- Visual Informatics Group