Simulation Based Synthesis of Planning Logic for Autonomous Unmanned Sea Surface Vehicles
S. K. Gupta, B. Kavetsky, S. Lubard, M. Schwartz, P. Svec, and A. Thakur
This work is supported by the Office of Naval Research (ONR).
Autonomy, Unmanned Sea Surface Vehicle, Planning logic, Evolutionary Robotics
Technological advances in unmanned sea surface vehicles (USVs) have enabled the unmanned boats to be involved in search, rescue, recovery, surveillance, and reconnaissance applications. Currently, most USVs are semiautonomous. This means that the way-points programmed into these semiautonomous USVs are initially determined by human navigators and the built-in USV navigation planners ensure safe movement between the way-points. Some also compute new way-points or employ a few emergency actions, such as abort and stop, in response to fault conditions.
A major issue in the development of increased autonomy for robotic vehicles such as USVs is the time and expense of developing the software necessary to handle a large variety of missions and all the variations in the encountered environments. The high cost of the development is mainly due to challenging implementation of certain parts of the missions that involve multiple agents in highly dynamic environments.
Over the last two decades, the advent of information technology has significantly influenced all facets of the engineering practice. So a natural question is Ė what role can computers play in the innovation and discovery process? Recent advances in the high fidelity simulations enable us to do an accurate analysis of proposed solution. The advent of low-cost, high-performance computing architectures enables us to explore a very large number of solutions in a short period of time. Advances in machine learning and procedural representations allow us to automatically generate complex candidate solutions. Hence, our premise is that high fidelity simulations together with advanced machine learning techniques can now facilitate innovation and discovery process.
The objective of this project is to develop a system which can automatically generate planning logic for autonomous operation of USSV-HTF for a scenario of interest. Specifically, we are interested in the following:
- Automatically generate a planning logic that leads to the USSV-HTF autonomous behaviors.
- Demonstrate that automatically generating logic leads to cost savings.
- Represent logic in a symbolic form which enables verification.
Overview of Approach
The basic idea behind our approach is as follows. The USV explores the virtual environment by randomly trying different moves. USV moves are simulated in the virtual environment and evaluated based on their ability to make progress toward the mission goal. If a successful action is identified as a part of the random exploration, then this action will be integrated into the logic driving the USV. We anticipate that there may be portions of the mission, where trial and error alone will not be adequate to discover the right decision rule. In such cases, two additional approaches are utilized to make progress in acquiring the right logic. The first approach involves seeding the system with the logic employed by humans to solve a challenging task. The second approach is to restrict the action space based on some type of feasibility criteria. Currently, our focus is specifically on automated generation of a human-readable planning logic used for blocking the advancement of an intruder boat towards a valuable target. For the machine learning process to successfully produce a human-competitive blocking logic, this specific task requires a suitable human-competitive adversary serving as a performance test for the USVís blocking capabilities.
Development of a physics-based meta-model: High fidelity simulation of the USV is time consuming and cannot be used for discovering decision rules or trees used in planning. We are currently in the process of developing a meta-model by conducting off-line simulations of the USV in the sea. This simulation accounts for wave and USV interactions. The meta-model provides information about turning radius, steady state velocity, and acceleration as a function of rudder angle and throttle position.
Development of mission planning system: Evolutionary Robotics (ER) is a methodology that uses evolutionary algorithms to automatically synthesize controllers and body configuration for autonomous robots. In most applications the evolution happens without human interference and in close interaction with the environment. We have developed a mission planning system with an evolutionary module for generating planning decision trees as the main component. Particularly, we utilized the genetic programming (GP) as one of the robust evolutionary techniques to automatically generate decision trees expressing blocking logic for the USV. Instead of automatically generating a program composed of low-level controller actions (steer left/right, go straight), we generate a decision tree that consists of high-level controllers as building blocks (go in front of the intruder, etc.) together with conditionals and other program constructs.
Development of USV simulation environment: In order to combine all the elements of the project into a cohesive system, we have designed a USV simulation environment. The USV simulation environment integrates various components of the project into a complete simulation system and acts as a simulation platform for the evolutionary module. One of the components of the USV simulation environment is the virtual environment (VE) based simulator which serves as an emulator of the real USV environment and contains gaming logic which allows human players to play against each other or against the computer. The game logic is responsible for the rules of the game, game logging and replay, boat behaviors, and scoring.
Evaluation of planning logic performance: Deployment of autonomous USVs in critical missions requires that the performance of the autonomous system matches with that of a remote controlled vehicle. Therefore, we have started an effort to assess the performance of automatically generated blocking logic compared to blocking maneuvers exhibited by human operators. We have used the USV simulation environment to compare the automatically discovered decision trees representing the blocking logic to the strategies used by human players.
M. Schwartz, P. Svec, A. Thakur, and S. K. Gupta, "Evaluation of Automatically Generated Reactive Planning Logic for Unmanned Surface Vehicles," in Performance Metrics for Intelligent Systems Workshop Gaithersburg, Maryland, 2009.
P. Svec and S. K. Gupta, "Competitive Co-evolution of High-Level Blocking Controllers for Unmanned Surface Vehicles," in Exploring New Horizons in Evolutionary Design of Robots Workshop (IEEE IROS 2009) St. Louis, Missouri, 2009.
For additional information please contact:
Dr. Satyandra K. Gupta
Department of Mechanical Engineering and Institute for Systems Research
3143 Martin Hall
University of Maryland
College Park, MD-20742