Search for: All records

The ability of quadrupedal robots to follow commanded velocities is important for navigating in constrained environments such as homes and warehouses. This paper presents a simple, scalable approach to realize high fidelity speed regulation and demonstrates its efficacy on a quadrupedal robot. Using analytical inverse kinematics and gravity compensation, a task-level controller calculates joint torques based on the prescribed motion of the torso. Due to filtering and feedback gains in this controller, there is an error in tracking the velocity. To ensure scalability, these errors are corrected at the time scale of a step using a Poincar´e map (a mapping of states and control between consecutive steps). A data-driven approach is used to identify a decoupled Poincar´e map, and to correct for the tracking error in simulation. However, due to model imperfections, the simulation-derived Poincar´e map-based controller leads to tracking errors on hardware. Three modeling approaches – a polynomial, a Gaussian process, and a neural network – are used to identify a correction to the simulation-based Poincar´e map and to reduce the tracking error on hardware. The advantages of our approach are the computational simplicity of the task-level controller (uses analytical computations and avoids numerical searches) and scalability of the sim-to-real transfer (use of low-dimensional Poincar´e map for sim-to-real transfer). A video is in this shortened link: http://tiny.cc/humanoids23 

In the oil and gas industry, exploration is largely dependent on the study of the subsurface hundreds or thousands of feet below. Most of the data used for this purpose is collected using borehole logging tools. Although sophisticated, these tools are limited as to how precisely they can measure the subsurface in terms of vertical resolution. There is one method of studying the subsurface that provides unlimited vertical resolution – core samples. Although core samples provide scientists the opportunity to generate a full, continuous data set, lab analysis work is normally done at one-foot intervals, as anything more would be prohibitively expensive. This means at best, a representative data set is generated. However, if the subsurface is not homogeneous, it is difficult to generate a representative data set with lab analysis done at one-foot intervals. This is a void that artificial intelligence can fill. More specifically, a properly trained neural network can analyze high-resolution core images continuously from top to bottom and generate a continuous analysis. It is also important to note that geologic interpretation tied to core analysis can introduce human error and subjectivity. Here too, a properly trained neural network can generate results with extreme levels of accuracy and precision. One core analysis expert believes that core analysis done manually is flawed about 70% of the time. This flawed analysis can result from lack of experience and or a lack of knowledge of the geologic formation. We are not the first to attempt to analyze core samples with vision algorithms. A group of Stanford researchers used micro-computed tomography (micro-CT) and Scanning Electron Microscopy (SEM) images of core samples to characterize the porous media. While promising, SEM and micro-CT imaging is expensive, and more importantly it is not a standard practice in the oil and gas industry to collect these types of images, making these images rare. One other work applied convolutional neural networks to a GIS based regional saturation system, but our work is significantly different. It is well known that training a neural network requires abundant data, thankfully with the method of core analysis we are proposing that will not be a problem. Through industrial partnerships we’ve obtained hundreds to thousands of core images sufficient to train a neural network, as well as core interpretations tied to those images coming from a core analysis expert with over 40 years of experience. We are the first to propose automatic hydrocarbon saturation as well as lithology prediction from core slab images. We propose the use of convolutional neural networks to analyze core samples at a single site. We plan to conduct experiments using a variety of neural networks to determine the best practices, and explore how such a service can be offered to the industry via the software-as-a-service paradigm. In the past, automated analysis through core slab images has not been possible simply because images of the required resolution were not common, but that has changed. If implemented successfully, this proposed method could become the new standard for core evaluation. 

The North Icelandic Jet (NIJ) is an important source of dense water to the overflow plume passing through Denmark Strait. The properties, structure, and transport of the NIJ are investigated for the first time along its entire pathway following the continental slope north of Iceland, using 13 hydrographic/velocity surveys of high spatial resolution conducted between 2004 and 2018. The comprehensive dataset reveals that the current originates northeast of Iceland and increases in volume transport by roughly 0.4 Sv (1 Sv ≡ 10 6 m 3 s −1 ) per 100 km until 300 km upstream of Denmark Strait, at which point the highest transport is reached. The bulk of the NIJ transport is confined to a small area in Θ– S space centered near −0.29° ± 0.16°C in Conservative Temperature and 35.075 ± 0.006 g kg −1 in Absolute Salinity. While the hydrographic properties of this transport mode are not significantly modified along the NIJ’s pathway, the transport estimates vary considerably between and within the surveys. Neither a clear seasonal signal nor a consistent link to atmospheric forcing was found, but barotropic and/or baroclinic instability is likely active in the current. The NIJ displays a double-core structure in roughly 50% of the occupations, with the two cores centered at the 600- and 800-m isobaths, respectively. The transport of overflow water 300 km upstream of Denmark Strait exceeds 1.8 ± 0.3 Sv, which is substantially larger than estimates from a year-long mooring array and hydrographic/velocity surveys closer to the strait, where the NIJ merges with the separated East Greenland Current. This implies a more substantial contribution of the NIJ to the Denmark Strait overflow plume than previously envisaged. 

Abstract The structure, transport, and seasonal variability of the West Greenland boundary current system near Cape Farewell are investigated using a high-resolution mooring array deployed from 2014 to 2018. The boundary current system is comprised of three components: the West Greenland Coastal Current, which advects cold and fresh Upper Polar Water (UPW); the West Greenland Current, which transports warm and salty Irminger Water (IW) along the upper slope and UPW at the surface; and the Deep Western Boundary Current, which advects dense overflow waters. Labrador Sea Water (LSW) is prevalent at the seaward side of the array within an offshore recirculation gyre and at the base of the West Greenland Current. The 4-yr mean transport of the full boundary current system is 31.1 ± 7.4 Sv (1 Sv ≡ 10 6 m 3 s −1 ), with no clear seasonal signal. However, the individual water mass components exhibit seasonal cycles in hydrographic properties and transport. LSW penetrates the boundary current locally, through entrainment/mixing from the adjacent recirculation gyre, and also enters the current upstream in the Irminger Sea. IW is modified through air–sea interaction during winter along the length of its trajectory around the Irminger Sea, which converts some of the water to LSW. This, together with the seasonal increase in LSW entering the current, results in an anticorrelation in transport between these two water masses. The seasonality in UPW transport can be explained by remote wind forcing and subsequent adjustment via coastal trapped waves. Our results provide the first quantitatively robust observational description of the boundary current in the eastern Labrador Sea. 

Abstract Solute exclusion during sea ice formation is a potentially important contributor to the Arctic Ocean inorganic carbon cycle that could increase as ice cover diminishes. When ice forms, solutes are excluded from the ice matrix, creating a brine that includes dissolved inorganic carbon (DIC) and total alkalinity (A_T). The brine sinks, potentially exporting DIC andA_Tto deeper water. This phenomenon has rarely been observed, however. In this manuscript, we examine a ~1 yearpCO₂mooring time series where a ~35‐μatm increase inpCO₂was observed in the mixed layer during the ice formation period, corresponding to a simultaneous increase in salinity from 27.2 to 28.5. Using salinity and ice based mass balances, we show that most of the observed increases can be attributed to solute exclusion during ice formation. The resultingpCO₂is sensitive to the ratio ofA_Tand DIC retained in the ice and the mixed layer depth, which controls dilution of the ice‐derivedA_Tand DIC. In the Canada Basin, of the ~92 μmol/kg increase in DIC, 17 μmol/kg was taken up by biological production and the remainder was trapped between the halocline and the summer stratified surface layer. Although not observed before the mooring was recovered, this inorganic carbon was likely later entrained with surface water, increasing thepCO₂at the surface. It is probable that inorganic carbon exclusion during ice formation will have an increasingly important influence on DIC andpCO₂in the surface of the Arctic Ocean as seasonal ice production and wind‐driven mixing increase with diminishing ice cover.