Autonomous Path Planning for Information Gathering In Lethally Hostile Environments with Severely Limited Communication
Abstract
The Navy may operate in environments that are both lethally hostile and severely constrained with respect to communication. We propose fundamental research into planning algorithms for autonomous agents that operate in such environments; in particular, to gather information from the environment in the presence of hazards, vehicle losses, and severely limited communication– while using information theoretic measures of performance. We are motivated by the following scenario: Autonomous agents search for human survivors (targets) in a hazardous environment, but wireless communication is either prohibited or severely constrained. As agents gather information, they must periodically return to base to ensure that the information they have gathered is delivered to human commanders. The hazards are lethal (they can can destroy the agents), and are effectively invisible—because the only way to observe a hazard is to be destroyed by it. Information gathered by a particular agent, but not yet delivered to base, is lost in the event that the agent is destroyed. Although hazards are invisible, information about hazard locations can be inferred by remembering the path an agent plans to take, and then observing whether or not the agent survives its journey. Agents are less likely to return from paths containing adversaries. Thus, agent path survival can be used as a noisy hazards sensor, and our evolving belief about hazard locations can be used to inform future search plans. Preliminary work to address the feasibility of this idea began when the PI was a postdoc at NRL and has continued at the University of Maryland using start-up funds. The preliminary investigation considered (only) the simple case that communication is impossible, and was presented at the Workshop on the Algorithm foundation of Robotics (WAFR). The new fundamental research we propose is designed to build upon our preliminary work in three novel ways: Thrust I: Research into the more general scenario where noisy communication is possible in the environment, but varies in reliability according to a distribution that is unknown at the beginning of the mission. By observing the success or failure of communications from the field, we hope to (i) infer the communication qualities of the environment in parallel to both (ii) searching for targets and (iii) inferring hazard locations based on agent survival. Knowledge gained about the environment’s communication properties can be leveraged to enable better (more informative) paths, since more information may be opportunistically transmitted to base, on average. This will be especially helpful in scenarios where agents are likely to be destroyed. Thrust II: Research into the problem variant in which there is a correlation between hazard locations and target locations. Our preliminary work does not account for correlations, and there are many scenarios of possible interest to the Navy in which correlations exist. For example, if targets are human pilots that need to be rescued because their aircraft have been destroyed by hazards. Thrust III: Research into the case where the probability that a hazard destroys a target changes as a function of time. For example, hazard lethality increases the longer an agent is near a hazard. This more accurately reflects s cenarios o f p ossible i nterest t o t he N avy: a dversaries may become more lethal given more time to observe our agent’s movement, and hazards such as radiation have cumulative effects. This research will potentially benefit Naval missions that require autonomous information gathering in hazardous and communication constrained environments. In particular, in cases where we have cheap, expendable, autonomous agents; and loss rates are expected to be high. Example missions include: search and rescue, surveillance, target localization, hazard detection, and intelligence gathering and/or scouting missions.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 20, 2020
- Source ID
- N000142012712
Entities
People
- Michael Otte
Organizations
- Office of Naval Research
- United States Navy
- University of Maryland