-
ReACT: Reinforcement Learning for Controller Parametrization using B-Spline Geometries
Authors:
Thomas Rudolf,
Daniel Flögel,
Tobias Schürmann,
Simon Süß,
Stefan Schwab,
Sören Hohmann
Abstract:
Robust and performant controllers are essential for industrial applications. However, deriving controller parameters for complex and nonlinear systems is challenging and time-consuming. To facilitate automatic controller parametrization, this work presents a novel approach using deep reinforcement learning (DRL) with N-dimensional B-spline geometries (BSGs). We focus on the control of parameter-va…
▽ More
Robust and performant controllers are essential for industrial applications. However, deriving controller parameters for complex and nonlinear systems is challenging and time-consuming. To facilitate automatic controller parametrization, this work presents a novel approach using deep reinforcement learning (DRL) with N-dimensional B-spline geometries (BSGs). We focus on the control of parameter-variant systems, a class of systems with complex behavior which depends on the operating conditions. For this system class, gain-scheduling control structures are widely used in applications across industries due to well-known design principles. Facilitating the expensive controller parametrization task regarding these control structures, we deploy an DRL agent. Based on control system observations, the agent autonomously decides how to adapt the controller parameters. We make the adaptation process more efficient by introducing BSGs to map the controller parameters which may depend on numerous operating conditions. To preprocess time-series data and extract a fixed-length feature vector, we use a long short-term memory (LSTM) neural networks. Furthermore, this work contributes actor regularizations that are relevant to real-world environments which differ from training. Accordingly, we apply dropout layer normalization to the actor and critic networks of the truncated quantile critic (TQC) algorithm. To show our approach's working principle and effectiveness, we train and evaluate the DRL agent on the parametrization task of an industrial control structure with parameter lookup tables.
△ Less
Submitted 10 January, 2024;
originally announced January 2024.
-
Real-time Dynamic Object Detection for Autonomous Driving using Prior 3D-Maps
Authors:
B Ravi Kiran,
Luis Roldão,
Benat Irastorza,
Renzo Verastegui,
Sebastian Suss,
Senthil Yogamani,
Victor Talpaert,
Alexandre Lepoutre,
Guillaume Trehard
Abstract:
Lidar has become an essential sensor for autonomous driving as it provides reliable depth estimation. Lidar is also the primary sensor used in building 3D maps which can be used even in the case of low-cost systems which do not use Lidar. Computation on Lidar point clouds is intensive as it requires processing of millions of points per second. Additionally there are many subsequent tasks such as c…
▽ More
Lidar has become an essential sensor for autonomous driving as it provides reliable depth estimation. Lidar is also the primary sensor used in building 3D maps which can be used even in the case of low-cost systems which do not use Lidar. Computation on Lidar point clouds is intensive as it requires processing of millions of points per second. Additionally there are many subsequent tasks such as clustering, detection, tracking and classification which makes real-time execution challenging. In this paper, we discuss real-time dynamic object detection algorithms which leverages previously mapped Lidar point clouds to reduce processing. The prior 3D maps provide a static background model and we formulate dynamic object detection as a background subtraction problem. Computation and modeling challenges in the map** and online execution pipeline are described. We propose a rejection cascade architecture to subtract road regions and other 3D regions separately. We implemented an initial version of our proposed algorithm and evaluated the accuracy on CARLA simulator.
△ Less
Submitted 5 July, 2019; v1 submitted 28 September, 2018;
originally announced September 2018.
-
Morphology, dynamics and plasma parameters of plumes and inter-plume regions in solar coronal holes
Authors:
K. Wilhelm,
L. Abbo,
F. Auchere,
N. Barbey,
L. Feng,
A. H. Gabriel,
S. Giordano,
S. Imada,
A. Llebaria,
W. H. Matthaeus,
G. Poletto,
N. -E. Raouafi,
S. T. Suess,
L. Teriaca,
Y. -M. Wang
Abstract:
Coronal plumes, which extend from solar coronal holes (CH) into the high corona and - possibly - into the solar wind (SW), can now continuously be studied with modern telescopes and spectrometers on spacecraft, in addition to investigations from the ground, in particular, during total eclipses. Despite the large amount of data available on these prominent features and related phenomena, many quest…
▽ More
Coronal plumes, which extend from solar coronal holes (CH) into the high corona and - possibly - into the solar wind (SW), can now continuously be studied with modern telescopes and spectrometers on spacecraft, in addition to investigations from the ground, in particular, during total eclipses. Despite the large amount of data available on these prominent features and related phenomena, many questions remained unanswered as to their generation and relative contributions to the high-speed streams emanating from CHs. An understanding of the processes of plume formation and evolution requires a better knowledge of the physical conditions at the base of CHs, in plumes and in the surrounding inter-plume regions (IPR). More specifically, information is needed on the magnetic field configuration, the electron densities and temperatures, effective ion temperatures, non-thermal motions, plume cross-sections relative to the size of a CH, the plasma bulk speeds, as well as any plume signatures in the SW. In spring 2007, the authors proposed a study on "Structure and dynamics of coronal plumes and inter-plume regions in solar coronal holes" to the International Space Science Institute (ISSI) in Bern to clarify some of these aspects by considering relevant observations and the extensive literature. This review summarizes the results and conclusions of the study. Stereoscopic observations allowed us to include three-dimensional reconstructions of plumes. Multi-instrument investigations carried out during several campaigns led to progress in some areas, such as plasma densities, temperatures, plume structure and the relation to other solar phenomena, but not all questions could be answered concerning the details of plume generation process(es) and interaction with the SW.
△ Less
Submitted 23 March, 2011;
originally announced March 2011.
-
Ultraviolet Coronagraph Spectroscopy: A Key Capability for Understanding the Physics of Solar Wind Acceleration
Authors:
S. R. Cranmer,
J. L. Kohl,
D. Alexander,
A. Bhattacharjee,
B. A. Breech,
N. S. Brickhouse,
B. D. G. Chandran,
A. K. Dupree,
R. Esser,
S. P. Gary,
J. V. Hollweg,
P. A. Isenberg,
S. W. Kahler,
Y. -K. Ko,
J. M. Laming,
E. Landi,
W. H. Matthaeus,
N. A. Murphy,
S. Oughton,
J. C. Raymond,
D. B. Reisenfeld,
S. T. Suess,
A. A. van Ballegooijen,
B. E. Wood
Abstract:
Understanding the physical processes responsible for accelerating the solar wind requires detailed measurements of the collisionless plasma in the extended solar corona. Some key clues about these processes have come from instruments that combine the power of an ultraviolet (UV) spectrometer with an occulted telescope. This combination enables measurements of ion emission lines far from the bright…
▽ More
Understanding the physical processes responsible for accelerating the solar wind requires detailed measurements of the collisionless plasma in the extended solar corona. Some key clues about these processes have come from instruments that combine the power of an ultraviolet (UV) spectrometer with an occulted telescope. This combination enables measurements of ion emission lines far from the bright solar disk, where most of the solar wind acceleration occurs. Although the UVCS instrument on SOHO made several key discoveries, many questions remain unanswered because its capabilities were limited. This white paper summarizes these past achievements and also describes what can be accomplished with next-generation instrumentation of this kind.
△ Less
Submitted 10 November, 2010;
originally announced November 2010.
-
Morphology and density of post-CME current sheets
Authors:
B. Vršnak,
G. Poletto,
E. Vujić,
A. Vourlidas,
Y. -K. Ko,
J. C. Raymond,
A. Ciaravella,
T. Žic,
D. F. Webb,
A. Bemporad,
F. Landini,
G. Schettino,
C. Jacobs,
S. T. Suess
Abstract:
Eruption of a coronal mass ejection (CME) drags and "opens" the coronal magnetic field, presumably leading to the formation of a large-scale current sheet and the field relaxation by magnetic reconnection. We analyze physical characteristics of ray-like coronal features formed in the aftermath of CMEs, to check if the interpretation of this phenomenon in terms of reconnecting current sheet is co…
▽ More
Eruption of a coronal mass ejection (CME) drags and "opens" the coronal magnetic field, presumably leading to the formation of a large-scale current sheet and the field relaxation by magnetic reconnection. We analyze physical characteristics of ray-like coronal features formed in the aftermath of CMEs, to check if the interpretation of this phenomenon in terms of reconnecting current sheet is consistent with the observations. The study is focused on measurements of the ray width, density excess, and coronal velocity field as a function of the radial distance. The morphology of rays indicates that they occur as a consequence of Petschek-like reconnection in the large scale current sheet formed in the wake of CME. The hypothesis is supported by the flow pattern, often showing outflows along the ray, and sometimes also inflows into the ray. The inferred inflow velocities range from 3 to 30 km s$^{-1}$, consistent with the narrow opening-angle of rays, adding up to a few degrees. The density of rays is an order of magnitude larger than in the ambient corona. The density-excess measurements are compared with the results of the analytical model in which the Petschek-like reconnection geometry is applied to the vertical current sheet, taking into account the decrease of the external coronal density and magnetic field with height. The model results are consistent with the observations, revealing that the main cause of the density excess in rays is a transport of the dense plasma from lower to larger heights by the reconnection outflow.
△ Less
Submitted 23 February, 2009;
originally announced February 2009.
