Christopher Goodin

A Real-Time, Interactive Simulation Environment for Unmanned Ground Vehicles: The Autonomous Navigation Virtual Environment Laboratory (ANVEL)

Modeling and simulation tools have become an integral part of modern engineering processes. In particular, accurate and efficient simulation tools are critical for the design, development, and testing of autonomous unmanned ground vehicles (UGVs). However, because of the complexity of the problem, many UGV simulators are computationally intensive, require expensive hardware to run, and are often not interactive or real-time. Those simulation environments that do provide users with some degree of interactivity and real-time or faster performance gain these features at the sacrifice of simulation fidelity, and these products often provide inadequate results. A new simulation environment for UGV design and development, called the Autonomous Navigation Virtual Environment Laboratory (ANVEL), has been created to address the need for a real-time, interactive, physics-based simulation environment for UGVs. ANVEL is able to meet this need by fusing readily available, off-the-shelf video game technology with high-fidelity, physics-based models. This paper presents the methodology used in developing ANVEL, an example use of ANVEL for development and testing of an autonomous UGV, and plans for the future development.

High fidelity sensor simulations for the virtual autonomous navigation environment

The Virtual Autonomous Navigation Environment (VANE) is a high-fidelity simulation environment for ground robotics. Physicsbased realism is the primary goal of the VANE. The VANE simulation incorporates realistic lighting, vehicle-terrain interaction, environmental attributions, and sensors. The sensor models, including camera, laser ranging, and GPS, are the focus of this work. These sensor models were designed to incorporate both internal (electronic) and external (environment) noise in order to produce a sensor output that closely matches that produced in real-world environments. This sensor output will allow roboticists to use simulation further into the development and debugging process before exposing robots to field conditions.

High-fidelity physics-based simulation of a UGV reconnaissance mission in a complex urban environment

Physics-based simulations of autonomous unmanned ground vehicles (UGV) present unique challenges and advantages compared to real-time simulations with lower-fidelity models. We have created a high-fidelity simulation environment, called the Virtual Autonomous Navigation Environment (VANE), to perform physics-based simulations of UGV. To highlight the capabilities of the VANE, we recently completed a simulation of a robot performing a reconnaissance mission in a typical Middle Eastern town. The result of the experiment demonstrated the need for physics-based simulation for certain circumstances such as LADAR returns from razor wire and GPS dropout and dilution of precision in urban canyons.

The Need for High-Fidelity Robotics Sensor Models

Simulations provide a safe, controlled setting for testing and are therefore ideal for rapidly developing and testing autonomous mobile robot behaviors. However, algorithms for mobile robots are notorious for transitioning poorly from simulations to fielded platforms. The difficulty can in part be attributed to the use of simplistic sensor models that do not recreate important phenomena that affect autonomous navigation. The differences between the output of simple sensor models and true sensors are highlighted using results from a field test exercise with the National Robotics Engineering Centers Crusher vehicle. The Crusher was manually driven through an area consisting of a mix of small vegetation, rocks, and hay bales. LIDAR sensor data was collected along the path traveled and used to construct a model of the area. LIDAR data were simulated using a simple point-intersection model for a second, independent path. Cost maps were generated by the Crusher autonomy system using both the real-world and simulated sensor data. The comparison of these cost maps shows consistencies on most solid, large geometry surfaces such as the ground, but discrepancies around vegetation indicate that higher fidelity models are required to truly capture the complex interactions of the sensors with complex objects.

A probabilistic model for simulating the effect of airborne dust on ground-based LIDAR

Field and laboratory measurements of Light Detection and Ranging (LIDAR) sensor interactions with dust have been performed for two types of common ground-based LIDAR sensors. A strong correlation (R2 > 0.95) between the probability for a return from the dust and the optical depth of the dust was found in the analysis. Based on the experimental correlation, a probabilistic model for LIDAR interactions with dust was developed and verified in field experiments. Finally, the model was integrated into a high-fidelity ray-tracing simulation of LIDAR systems.

NIR sensitivity analysis with the VANE

Near infrared (NIR) cameras, with peak sensitivity around 905-nm wavelengths, are increasingly used in object detection applications such as pedestrian detection, occupant detection in vehicles, and vehicle detection. In this work, we present the results of simulated sensitivity analysis for object detection with NIR cameras. The analysis was conducted using high performance computing (HPC) to determine the environmental effects on object detection in different terrains and environmental conditions. The Virtual Autonomous Navigation Environment (VANE) was used to simulate highresolution models for environment, terrain, vehicles, and sensors. In the experiment, an active fiducial marker was attached to the rear bumper of a vehicle. The camera was mounted on a following vehicle that trailed at varying standoff distances. Three different terrain conditions (rural, urban, and forest), two environmental conditions (clear and hazy), three different times of day (morning, noon, and evening), and six different standoff distances were used to perform the sensor sensitivity analysis. The NIR camera that was used for the simulation is the DMK firewire monochrome on a pan-tilt motor. Standoff distance was varied along with environment and environmental conditions to determine the critical failure points for the sensor. Feature matching was used to detect the markers in each frame of the simulation, and the percentage of frames in which one of the markers was detected was recorded. The standoff distance produced the biggest impact on the performance of the camera system, while the camera system was not sensitive to environment conditions.

Simulation of a multispectral, multicamera, off-road autonomous vehicle perception system with Virtual Autonomous Navigation Environment (VANE)

We present a case-study in using specialized, physics-based software for high-fidelity environment and electro-optical sensor modeling in order to produce simulated sensor data that can be used to train a multi-spectral perception system for unmanned ground vehicle navigation. This case-study used the Virtual Autonomous Navigation Environment (VANE) to simulate filtered, multi-spectral imaging sensors. The VANE utilizes ray-tracing and hyperspectral material properties to capture the sensor-environment interaction. In this study we focus on a digital scene of the ERDC test track in Vicksburg, MS that has extremely detailed representation of the vegetation and ground texture. The scene model is used to generate imagery that simulates the output of specialized terrain perception hardware developed by Southwest Research Institute, which consists of stereo pair of 3-channel cameras. The perception system utilizes stereo processing, the multi-spectral responses, and image texture features in order to create a 3-dimensional world model suitable for offroad vehicle navigation, providing depth information and an estimated terrain class label for every pixel by utilizing machine learning. While the process of training the perception system generally involves hand-labeling data collected through manned missions, the ability to generate data for certain environments and lighting conditions represents an enabling technology for deployment in new theaters. We demonstrate an initial capability to simulate data and train the perception system and present the results compared to the system trained with real-world data from the same location.

Analytic expressions for the black-sky and white-sky albedos of the cosine lobe model

The cosine lobe model is a bidirectional reflectance distribution function (BRDF) that is commonly used in computer graphics to model specular reflections. The model is both simple and physically plausible, but physical quantities such as albedo have not been related to the parameterization of the model. In this paper, analytic expressions for calculating the black-sky and white-sky albedos from the cosine lobe BRDF model with integer exponents will be derived, to the authors knowledge for the first time. These expressions for albedo can be used to place constraints on physics-based simulations of radiative transfer such as high-fidelity ray-tracing simulations.

Unmanned ground vehicle simulation with the Virtual Autonomous Navigation Environment

Unmanned and autonomous ground vehicles have the potential to revolutionize military and civilian navigation. Military vehicles, however, present unique challenges related to autonomous navigation that are not encountered in civilian applications. These include a high percentage of off-road navigation, navigation in hostile environments, navigation in GPS-denied environments, and navigation in urban environments where little data regarding road networks are available. These unique challenges require a deliberate approach for developing robust, reliable autonomous and unmanned systems that features extensive testing for performance and safety features in a wide variety of environments and conditions. In order to enable this approach, we developed a computational tool for simulating and predicting the performance of unmanned and autonomous ground vehicles in realistic environmental conditions. This tool, the Virtual Autonomous Navigation Environment, is discussed in this paper.

Computational intelligence-based optimization of maximally stable extremal region segmentation for object detection

Particle swarm optimization (PSO) and genetic algorithms (GAs) are two optimization techniques from the field of computational intelligence (CI) for search problems where a direct solution can not easily be obtained. One such problem is finding an optimal set of parameters for the maximally stable extremal region (MSER) algorithm to detect areas of interest in imagery. Specifically, this paper describes the design of a GA and PSO for optimizing MSER parameters to detect stop signs in imagery produced via simulation for use in an autonomous vehicle navigation system. Several additions to the GA and PSO are required to successfully detect stop signs in simulated images. These additions are a primary focus of this paper and include: the identification of an appropriate fitness function, the creation of a variable mutation operator for the GA, an anytime algorithm modification to allow the GA to compute a solution quickly, the addition of an exponential velocity decay function to the PSO, the addition of an ”execution best” omnipresent particle to the PSO, and the addition of an attractive force component to the PSO velocity update equation. Experimentation was performed with the GA using various combinations of selection, crossover, and mutation operators and experimentation was also performed with the PSO using various combinations of neighborhood topologies, swarm sizes, cognitive influence scalars, and social influence scalars. The results of both the GA and PSO optimized parameter sets are presented. This paper details the benefits and drawbacks of each algorithm in terms of detection accuracy, execution speed, and additions required to generate successful problem specific parameter sets.