ABSTRACT. This paper highlights the need for a new power model for skid-steered wheeled robots driving on loose terrains, and lays the groundwork to develop such a model. State-of-the-art power modeling assumes hard ground; under typical assumptions this predicts constant power consumption over a range of small turning radii where the inner wheels are rotating backwards. However, experimental results performed both in the field and in a controlled laboratory show that, in loose sandy soil, power is not in fact constant with respect to turning radius in this case. Power peaks in a newly identified range of turns where the inner wheels rotate backwards but are being dragged forward. Data shows higher motor torque and wheel sinkage in this range. A skidding wheel that is sunk into loose soil bulldozes a pile of sand; initial modeling of this phenomena reproduces the trend in additional power with respect to turning radius. As work on a full power model for loose terrain continues, this work identifies turning radii to avoid whenever possible in practice.

ABSTRACT. Exploring the water-rich surface of Europa is a major planetary exploration objective for the coming decades. Knowledge of the Europan terrain is relatively uncertain and therefore to perform mobile surface operations a versatile mobility system capable of adapting its locomotion is desirable. Our work summarizes preliminary efforts to evaluate a variety of wheel-on-limb mobility modes on Europan analogue terrain populated with unstructured salt-evaporite formations. A redundant wheel-on-limb quadruped, RoboSimian, has been adapted to perform traverses over unstructured terrain with obstacles up to 50 cm tall (1.6 times its wheel diameter). Multiple mobility modes that are combinations of wheeled and limbed motion primitives are proposed and evaluated over varying roughness terrain as a function of energy usage. In combination with these mobility modes, a mode-agnostic actively articulated suspension control scheme that maintains ground traction, regulates chassis orientation and prevents suspension saturation is presented. The controller uses an inverse kinematics solver that leverages pre-computed lookup-tables for each limb. This is shown to be simple yet effective in the field and is able to scale to a robot with 32 degrees-of-freedom.

ABSTRACT. Ice is the most valuable resource on the Moon. It exists only at the poles where shadows are extensive and drivable routes are short. Robot routes to reach this ice are tenuous. Sun-synchronous lunar polar routes offer order-of-magnitude greater duration and range, if such routes are achievable. Sun-synchrony is brittle in the sense that a rover must be at precisely scheduled time and place, so special localization techniques are warranted. Methods for terrain-based localization that work at equatorial regions are challenged at the lunar poles, where the grazing sunlight casts long shadows that obscure and change views over time. The shadows are shown here to accentuate craters as localization features. This paper presents a method that improves terrain registration at the poles of the Moon by probabilistically considering sensor and terrain uncertainty, and exploiting shadows as semantic features for localization. This method is validated and evaluated in simulated experiments.

ABSTRACT. The past few decades have seen tremendous progress in planetary exploration rovers and the most groundbreaking missions in space exploration history. Although the main focus has been toward single rover deployment, there has also been substantial research into swarm intelligence and multi-rover systems. This paper presents single and multi-rover path planning strategies for future planetary exploration missions. The main objective is to develop real-time Coverage Path Planning (CPP) solutions in unknown environments. The algorithms presented here are based on simple building blocks that use solely myopic sensing information to iteratively compute a motion cost function and decide their next move. Several single rover exploration methods are introduced and compared through simulation. The results show that complete coverage is not always possible in planetary exploration scenarios. Also, by slightly reducing the required percent of terrain to explore, the efficiency can be improved. In addition to this, two variants of multi-rover path planning algorithms are defined. A first method is a fully collaborative method where the rovers explore a region together and a second where the terrain is segmented and the rover operate separately from each other.

ABSTRACT. Long-term monitoring of natural environments raises significant challenges due to the strong perceptual aliasing in trees, bushes and shrubs. This paper reports on the multi-session localization and mapping of a small lake shore using an autonomous surface vessel equipped with a 2D lidar and a camera. Our publicly available dataset includes 130 autonomous surveys of the 1 km shoreline while recording lidar, GPS and image data. We build our globally consistent multi-session map using ICP at multiple scale. The end result is evaluated qualitatively by super-imposing all the lidar maps, and quantitatively by comparing images taken from the same pose at different times. The localization and mapping results, as well as the dataset of image pairs, are made available within our public dataset.

ABSTRACT. Autonomous coverage has tremendous importance for environmental surveying and exploration tasks performed on rivers both in terms of efficiency and data collection quality. Most surveys of rivers are performed manually using quite similar approaches. Using these practices to automate these processes improves the quality of survey operations. In addition to human expertise on the type of patterns, the coverage of a river can be optimized using the river meanders to determine the direction of coverage. In this work we use the implicit information on the speed of the water current, inferred from the curves of the river, to reduce the cost of coverage. We use autonomous surface vehicles (ASVs) to deploy the proposed methods and demonstrate the efficiency of our method. In addition we compare the proposed method with previous coverage techniques developed in our lab. When taking into account meanders the coverage time has been decreased in average by more than 20%. The deployments of the ASVs were performed on the Congaree River, SC, USA, and resulted in more than 27 km of total coverage trajectories.

ABSTRACT. The ability to navigate, search, and monitor dynamic marine environments such as ports, deltas, tributaries, and rivers presents several challenges to both human operated and autonomously operated surface vehicles. Human data collection and monitoring is overly taxing and inconsistent when faced with large coverage areas, disturbed environments, and potentially uninhabitable situations. In contrast, the same missions become achievable with autonomous surface vehicles (ASV) configured and capable of accurately maneuvering in such environments. The two dynamic factors that present formidable challenges to completing precise maneuvers in coastal and moving waters are currents and winds. In this work, we present novel and inexpensive methods for sensing these external forces, together with methods for accurately controlling an ASV in the presence of such external forces. The resulting platform is capable of deploying bathymetric and water quality monitoring sensors. Experimental results in the local lakes and rivers demonstrate the feasibility of the proposed approach.

ABSTRACT. This paper presents an approach for autonomous 3D semantic mapping of a coral reef. Current methods on coral reef health assessment are based on human observation. The proposed system organically joins a convolutional neural network (CNN) with a dense visual odometry approach and a correlation filter based tracker, KCF, to identify the different coral species detected during autonomous and manual data collection. In addition to the coral classification, the 3D position of each coral is identified producing a semantic map of the observed reef. Each coral is identified once, even when encountered at different times. The number of different coral species encountered during a single trajectory is reported. Furthermore, the shape and size of each coral are extracted from the dense reconstruction enabling the extraction of volumetric data for subsequent studies. Experimental results from the coral reefs of Barbados, utilizing different vehicles verify the robustness and accuracy of the proposed approach.

ABSTRACT. This paper presents a system for the intelligent search of shipwrecks using Autonomous Underwater Vehicles (AUVs). It introduces a machine learning approach to the automatic identification of potential archaeological sites from AUV-obtained side scan sonar (SSS) data. The site identification pipeline consists of a series of stages that set up for, run, and process the output of a convolutional neural network (CNN). To alleviate the issue of training data scarcity, i.e. the lack of SSS data that includes shipwrecks, and improve the performance at testing time, a data augmentation stage is included in the pipeline. In addition, edge detection and other traditional image processing feature extraction methods are used in parallel with the CNN to improve algorithmic performance. Experiments from two multi-deployment shipwreck search expeditions involving actual AUV deployments along the coast of Malta for data collection and processing demonstrate the pipeline’s usefulness. Results from these two field expeditions yielded a precision/recall of 29.34%/97.22% and 32.95%/80.39% respectively. Despite the poor precision, the pipeline filters out 99.79% of the area in data set A and 99.31% of the area in data set B.