Task-Aware Non-Gaussian Perception and Planning for Distributed Marine Autonomy
Abstract
This project will develop a new approach to task-aware and focussed perception and planning for Navy autonomous systems. Autonomous systems often encounter ambiguities that cannot be described with a unimodal Gaussian errordistribution; rather, the problem uncertainty is best represented by multiple modes of belief. Purely Gaussian representations often fail to capture the different types of uncertainty encountered by teams of robots performingdistributed perception and planning for long-term missions in difficult environments. Robust, long-term autonomy entails coping with large multi-sensor data streams while still being able to choose actions in real-time to resolve ambiguity and achieve mission objectives with minimal operator input. To address these questions in a new and fundamental way, we will explore two, interrelated thrusts: (1) distributed tractable non- Gaussian inference kernel Hilbert feature space (KHS) representations and (2) taskaware perception and planning using a novel focused inference framework for long-term data-rich resource-constrained domains. We will integrate and evaluate our algorithms using a variety of proxy platforms and sensors, tested at MIT in one or more of low-cost robot testing facilities.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 04, 2018
- Source ID
- N000141812832
Entities
People
- John J. Leonard
Organizations
- Massachusetts Institute of Technology
- Office of Naval Research
- United States Navy