Transactions of Society of Automotive Engineers of Japan
Online ISSN : 1883-0811
Print ISSN : 0287-8321
ISSN-L : 0287-8321
Current issue
Showing 1-32 articles out of 32 articles from the selected issue
Technical Paper
Research Paper
  • Tsubasa Matsura, Shota Ozawa, Riku Yamashita, Takanori Mieno, Fumio Sh ...
    2021 Volume 52 Issue 6 Pages 1173-1179
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    The prechamber combustion characteristics were studied using a constant volume combustion chamber to improve the efficiency of cogeneration natural gas engines. The torch flames generated by a prechamber were used to investigate the effect that a prechamber has on the main combustion. In our previous study, we observed the correlation between the torch flame and the main flame (which is a so-called “prechamber combustion”) as well as the knocking phenomena for various prechamber configurations using a rapid compression and expansion machine. In this study, we have investigated the effect of prechamber combustion on main chamber combustion characteristics using a constant volume combustion chamber, especially to study the ignitability of the main chamber for CH4, CH4 + H2, CH4 + C3H8. As a result, it was found that hydrogen addition improved the ignitability of the main chamber in the fuel lean side. In addition, there is a correlation between the experimental ignition delay time and the 0D calculation ignition delay time.
    Download PDF (718K)
  • Kazumasa Watanabe, Noboru Uchida
    2021 Volume 52 Issue 6 Pages 1180-1185
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    In order to achieve further improvement in the thermal efficiency of diesel combustion, it is necessary to identify the factors which suppress fuel-air mixture formation for achieving higher heat release rate and shorter combustion period. In this study, the effect of spray flame interference with adjacent flames and backflow of burnt gas on each spray flame were investigated by utilizing a bottom view transparent engine which can directly observe spray flame in the combustion chamber, and a numerical simulation. And it was revealed that those are the major factors for limiting the increase in heat release rate and prolonging after burning period with spatial constraints of the real engine.
    Download PDF (1695K)
  • - Effects of Ambient Conditions and Injection Pressure on Ignition Delay and Ignition Position -
    Kazuki Kawai, Yuki Wakai, Eriko Matsumura, Jiro Senda
    2021 Volume 52 Issue 6 Pages 1186-1191
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    The purpose of this study is to obtain a detailed understanding of the process from mixture formation to ignition in the diesel spray. In this paper, the effects of ambient temperature, ambient density, and injection pressure on the ignition delay and set-off length were measured using single-hole and three-holes nozzle with cone angle. The effects of adjacent sprays on the set-off length and ignition delay were also investigated by focusing on the backflow of ambient gas. In addition, a model was developed to predict the set-off length and ignition delay from the amount of ambient gas introduced into the spray.
    Download PDF (1262K)
  • Ryohei Noda, Yasuo Fujii, Hiroshi Mouri
    2021 Volume 52 Issue 6 Pages 1192-1197
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    Vehicles with a low steering gear ratio become sensitive to steering inputs of a specific frequency when driving at low speeds, which leads to poor riding comfort. It was clarified that this phenomenon occurs depending on the tire relaxation length and its mechanism. In a steer-by-wire vehicle, it was clarified in an actual vehicle experiment that it is possible to take measures against the above phenomenon by installing an appropriate filter between the steering wheel angle and the front wheel steering angle.
    Download PDF (1124K)
  • Tsuyoshi Katayama, Kenta Furusawa, Akira Takahashi, Takahiko Yoshino
    2021 Volume 52 Issue 6 Pages 1198-1203
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    It is well known that the straight running stability of a two-wheeled vehicle is affected by the frame flexibility. From the six types of frame flexibility models that have already been proposed, the effects of two models that greatly affect the weave mode, the torsional flexibility of the main frame and the torsional flexibility of the rear swing arm, are analyzed using a formulation method based on the eigenvector equation
    Download PDF (660K)
  • Takahiko Yoshino, Akira Takahashi, Kenta Furusawa, Tsuyoshi Katayama
    2021 Volume 52 Issue 6 Pages 1204-1211
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    In case a motorcycle is designed in a way that the weave mode is stabilized, the wobble mode tends to be unstable; similarly, when attempting to stabilize the wobble mode, the weave mode is likely to turn unstable. In this study, we investigated the effect of changing the extent of force acting on both modes using the energy flow method, and estimated the factors that affect the compatibility of the two modes. The results show that, in most cases, the front frame elements that show conflicting changes directly affect the coefficients of the equation of motion. Moreover, the stability of the compatible type often varies owing to changes in eigenvectors.
    Download PDF (875K)
  • - Prediction of Vibration Characteristics of Real Car Frame Based on Transfer Function Synthesis Method and Measurement -
    Ryo Ujima, Kazuki Murakami, Hiroki Nakamura, Toshiaki Kamo, Norikuni H ...
    2021 Volume 52 Issue 6 Pages 1212-1218
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    An estimation method based on the transfer function synthesis method for the vibration response of a real vehicle frame equipped with a micro vibration suppressor is developed and presented. Firstly, a measurement method of the excitation forces inputted to the frame through suspension is presented. Next, it is shown that the estimated and measured results of the vibration at the evaluation point agree in the accuracy of practical usage in a low frequency range between 4 and 20Hz sensitive to the ride quality. Lastly, a usefulness of this proposed method is verified from results of estimation accuracy and good relationship with the sensitive evaluation by professional test driver.
    Download PDF (1655K)
  • -Development of an Injury Prediction Algorithm Using Australian Accident Data-
    Kazuhiro Kubota, Tetsuya Nishimoto, Giulio Ponte
    2021 Volume 52 Issue 6 Pages 1219-1226
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    In this study, an injury prediction algorithm for vehicle occupants was developed based on the South Australian Traffic Accident Reporting System (TARS), a traffic accident database. The algorithm uses information that can be identified from the accident location or reported by bystanders (e.g., speed limit distance from the centre of Adelaide in SA, etc) as risk factors. The best combination of factors for injury prediction was selected from 15 items based on the Akaike Information Criteria (AIC). The algorithm had an under-triage rate of less than 10% for the serious injured and an overtriage rate of 45.1% for the minor injured. This algorithm can contribute to the reduction of the number of fatalities by helping Automatic Crash Notification (ACN) systems to operate as Advanced Automatic Crash Notification (AACN) systems, or to be used by emergency medical services in on-scene triage.
    Download PDF (507K)
  • Tetsuya Nishimoto, Ryota Mochizuki, Shigeru Tominaga, Tomoyuki Miyoshi ...
    2021 Volume 52 Issue 6 Pages 1227-1234
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    In this study, Version 2021 injury prediction algorithm, vehicle class was disaggregated into four categories; kei car (Japanese small car), small passenger car, medium passenger car and large passenger car. This was undertaken to enhance the prediction performance of Version 2015 and Version 2017 of the algorithm. The injury prediction algorithm Version 2021 was constructed using macro data from the Traffic Accident Database of the Institute for Traffic Accident Research and Analysis of Japan (ITARDA) and included 2,113,959 injured occupants in all four-wheel vehicle accidents that occurred in Japan from 2009 to 2018. In this study, a total of 44 regression models were constructed for driver, front passenger and rear seat passengers by vehicle class, crash direction and seat position. The risk factors were seatbelt use, occupant age and single/multiple impact scenarios. As an example using the new model, the new risk curve by reflecting the driver's seat position and by vehicle class for frontal collision of Version 2021.
    Download PDF (856K)
  • Goshi Sasaki, Trongmun Jiralerspong, Daisuke Asai
    2021 Volume 52 Issue 6 Pages 1235-1240
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    Unexpected operation made by the vehicle system are becoming difficult to analyze due to the sophistication of autonomous driving technology. As such, there is a growing demand to collect and classify images of various driving scenes using cloud platform, but this process requires a lot of workload. Therefore, it is necessary to develop a technology to automatically classify driving scenes based on the information of the ego vehicle’s surrounding situation. This paper proposes a method to accurately detect objects in front of the vehicle and estimate their distance using a perspective projection model by applying super-resolution and object detection techniques.
    Download PDF (934K)
  • Yuta Yokoyama, Kei Nagasaka, Idemitsu Masuda, Ichiro Kiriyama, Hirofum ...
    2021 Volume 52 Issue 6 Pages 1241-1246
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    We evaluate performance of Isogeometric analysis (IGA) for bending and buckling deformation of structural members. In automobile crash analysis, ensuring the accuracy of the acceleration, velocity and load in time series as well as the deformation behavior is important. For the above this consistency, accurate reproducing the bending and buckling of structural members are indispensable. In this study, we compute bending and buckling of structural members with IGA and validate its performance by comparing the results with conventional finite element analysis and experiments. In addition, we apply IGA to crash analysis of a car body.
    Download PDF (1609K)
  • Atsushi Itoh, Megumi Matsumoto, Naoki Hamamoto, Shinjiro Suzuki
    2021 Volume 52 Issue 6 Pages 1247-1253
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    This paper studied a shape optimization method using HPC (High Performance Computing) to obtain optimal designs which satisfy aerodynamics and underhood cooling airflow for a detailed full-vehicle model. The shape optimization workflow integrated with shape morphing, CFD (Computational Fluid Dynamics) and evolutionary computation was applied to a many-objective optimization for Cd and airflow velocities through heat exchangers. The result showed several optimal designs which all objectives are improved from the default.
    Download PDF (1396K)
  • Masatoshi Mikasa, Suguru Shiratori, Itsuhei Kohri, Hideaki Nagano, Ken ...
    2021 Volume 52 Issue 6 Pages 1254-1260
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    A series of three-dimensional time-dependent numerical simulations of the flow around the Ahmed Body were conducted for the case of a slant angle 𝜃= 29° and the range of the Reynolds number Re ⩽ 2000. The critical Reynolds number for the transition from the steady to the oscillatory flow was found to be Rec = 1751. The oscillatory flow appearing near Rec is of the single frequency oscillatory flow, which has the peak frequency of the Strouhal number St ≈ 0.07. For the higher Re the multiple frequency oscillatory flows are found, and the transition from the single to multiple oscillation can be predicted in the range 1800 < Re ⩽ 1900. The kinetic energy of the deviated oscillatory flow field is supplied from the time-averaged flow field strongly near the pair of the vertical vortices and near the rear wall of the body. As one possibility, the instability mechanism acting near the vertical vortices may have some similarity with known centrifugal instability.
    Download PDF (1443K)
  • — JSAE 2050 Challenge and Air Quality Estimation in 2050 —
    Tazuko Morikawa, Hiroyuki Yamada, Kotaro Tanaka, Shinichiro Okayama, Y ...
    2021 Volume 52 Issue 6 Pages 1261-1266
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    The domestic air pollutant emission for 2050 was estimated based on an 80% reduction of greenhouse gases by 2050. With the emission data and future estimation scenarios in mainland China, PM2.5 and O3 concentration were estimated by an air quality model: CMAQ (Community Multiscale Air Quality Modeling System). As a result, it was thought that the current atmospheric environmental quality standard would be achieved for PM2.5, and high-concentration photochemical O3 in summer would not occur. However, as the amount of decrease in O3 concentration in spring was small, it was considered that it would be difficult to achieve an air quality standard of 60 ppb.
    Download PDF (867K)
  • Hirofumi Sugiyama, Yasutake Haramiishi, Shigenobu Okazawa
    2021 Volume 52 Issue 6 Pages 1267-1272
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    This study performs a numerical simulation using a viscoplasticity model and considers internal fiber orientation for fiber reinforced plastics (FRP). The FRP, which the microstructure consists of resin and fiber, has a rate and fiber orientation dependency. However, it is difficult to predict the material parameters on the macro scale. Thus, we perform the material testing and decide the microstructure’s material parameters using the least squared method. Besides, we perform some macro numerical simulations that show the present method’s capabilities for the target material compared with the tensile testing results.
    Download PDF (558K)
  • Takeru Shiina, Yuki Matsuo, Souichiro Nishino, Eitaro Yukutake
    2021 Volume 52 Issue 6 Pages 1273-1278
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    In this study, 6061 aluminum alloy and 5052 aluminum alloy were welded homogeneously and dissimilarly by friction stirring, and the joint strength was evaluated by analyzing the detailed fracture mechanism. Furthermore, the effect of changing the stirring direction on the joint strength was also investigated. The joint strength was greater for upward stirring than downward stirring for both homogeneous and heterogeneous joints. The joint strength can be evaluated in a single sense by the hardness of the heat-affected zone (fracture zone), which is softened by the heat generated during friction stirring.
    Download PDF (1185K)
  • Yuto Ando, Souichiro Nishino, Ryo Tominaga, Naonori Ishii
    2021 Volume 52 Issue 6 Pages 1279-1284
    Published: 2021
    Released: October 22, 2021
    JOURNAL FREE ACCESS
    In the automotive industry, various types of materials are used to reduce weight, and it is necessary to join dissimilar metals. The plastic flow joining method is a simple technique for joining dissimilar metals with less thermal strain than welding. In this study, joint tests were conducted under three different conditions (pushing load, tip shape of jig and ring outer diameter) to investigate on joint strength. The strength prediction equation was proposed based on the analysis of the fracture mechanism on joint area during the tensile test. The joint strength increases as the pushing load increases, but excessive pushing load decreases the joint strength when the ring outer diameter is small. The tip shape of the jig had no effect on the joint strength under the conditions of this experiment. The joint strength can be predicted by the shear fracture of the material considering work hardening in plastic flow. The tensile strength of plastic worked material was calculated from Vickers hardness. As a result, the joint strength was determined from the shear strength of material and joint area. The results of calculated strength agreed well with the measured values.
    Download PDF (1052K)
  • Takashi Suzuki, Takashi Koyama, Kazuo Yamamori, Yuta Uematsu, Satoshi ...
    2021 Volume 52 Issue 6 Pages 1285-1291
    Published: 2021
    Released: November 25, 2021
    JOURNAL FREE ACCESS
    Reduction of base oil viscosity in engine oils can contribute to further fuel economy improvement of internal combustion engines. On the other hand, the lower viscosity base oil has a concern of oil consumption increase by evaporation. Although NOACK test has been utilized as volatility index related to the oil consumption, it cannot indicate the oil consumption sufficiently. In this study, a physical property of engine oil which can show the evaporation rate is investigated. As a result, saturated vapor pressure in the temperature range of cylinder bore showed good correlation to the oil consumption in actual engine operation.
    Download PDF (1102K)
  • Naoki Yoneya, Kengo Kumano, Atsushi Shimada, Yoshihiro Sukegawa, Kenta ...
    2021 Volume 52 Issue 6 Pages 1292-1297
    Published: 2021
    Released: November 25, 2021
    JOURNAL FREE ACCESS
    Pre-chamber jet ignition is one of the promising engine technologies for complying with the next pollutant emission regulation. In this paper, as a model-based design process, the investigation results of the improvement of gas-exchange in the pre-chamber and cooling performance based on CFD analysis will be reported. In addition, the single cylinder engine test results of the passive pre-chamber are introduced.
    Download PDF (1579K)
  • Hiroshi Hanabusa, Ryosuke Shiina, Suguru Fukuda, Noritaka Kimura
    2021 Volume 52 Issue 6 Pages 1298-1303
    Published: 2021
    Released: November 25, 2021
    JOURNAL FREE ACCESS
    Development of knocking mitigation technology is required to improve the thermal efficiency of gasoline engines. The purpose of this study was to analyze the effect of injection timing of a direct-injection gasoline engine on knocking and thermal efficiency and its mechanism, and to contribute to injector development and reduction of control adaptation man-hours. For the knocking mechanism, the contribution of combustion speed, in-cylinder gas temperature, and air-fuel-ratio distribution to mass fraction burned 50% crank angle were considered. In the indicated thermal efficiency, in addition to the contribution of mass burned fraction 50% crank angle, the contribution of unburned fuel loss and heat loss to the total heat release were considered.
    Download PDF (942K)
  • - Effect of coolant temperature and fuel injection timing on soot emissions -
    Jun Hashimoto, Kimitoshi Tanoue, Sangmyeong Kim, Tatsuya Kuboyama, Yas ...
    2021 Volume 52 Issue 6 Pages 1304-1310
    Published: 2021
    Released: November 25, 2021
    JOURNAL FREE ACCESS
    Direct-injection spark-ignition (DISI) engines have high thermal efficiency; however, they also show high soot emissions under cold-start conditions. It is also well known that the fuel injection timing (SOI) has a significant effect upon soot emissions. Therefore, it is important to reproduce effects of the coolant temperature and SOI on soot emissions in design calculations. In this study, numerical calculations of soot emissions from a DISI engine were performed by using a compact-size soot-formation model previously proposed by authors. We discussed the reproducibility of soot emissions by changing the coolant temperature and SOI. In addition, the experimentally observed tendency that soot emission slightly increase under fuel injection timings near the bottom dead center were discussed based on the calculated results.
    Download PDF (873K)
  • - Learning Based on State Assuming a Real Vehicle-
    Kazuki Ogawa, Tatsuhito Aihara
    2021 Volume 52 Issue 6 Pages 1311-1316
    Published: 2021
    Released: November 25, 2021
    JOURNAL FREE ACCESS
    Previous paper aims to apply the deep reinforcement learning to the shift control of the EV two-speed transmission. In this study, we perform learning and verification based on the state quantity assuming the actual vehicle, and clarify that deep reinforcement learning is also useful for the state quantity of the actual vehicle. In order to reproduce the actual vehicle condition, we have developed a simulation model that does not monitor the clutch transmission torque. Next, the robust controller by learning under multiple conditions is verified whether it is possible to develop a robust controller.
    Download PDF (1017K)
  • Hironori Teramura, Toshiro Matsuda, Yuki Sakai, Katsuya Nakano
    2021 Volume 52 Issue 6 Pages 1317-1322
    Published: 2021
    Released: November 25, 2021
    JOURNAL FREE ACCESS
    We have achieved performance design for affordable EV bus which includes high power motor system performance estimation, and fuel economy simulation using numerous route bus operation data in Yokohama city, additionally, we have completed demonstration test in FY2020 to verify EV performance which has been confirmed as good.
    Download PDF (1560K)
  • Takahiro Kawaguchi, Takahiro Ohyama, Ichiro Maruta, Kensuke Osamura, J ...
    2021 Volume 52 Issue 6 Pages 1323-1328
    Published: 2021
    Released: November 25, 2021
    JOURNAL FREE ACCESS
    This paper proposes a recursive estimation method of the effective resistance, which is a feature used to estimate the state of power of a battery. In the proposed method, a model of the internal impedance is constructed with a system identification technique, and the effective resistance is calculated from the model. This paper considers what structure of the model is appropriate for the calculation of the effective resistance and proposes using a model named μ-Markov model. This structure helps to reduce the number of estimated parameters and computational burden. The model is extended to be applied for data that have an offset. The parameters of the model are estimated by the recursive leastsquares with a forgetting factor. The effectiveness of the proposed method is verified through a numerical simulation and an experiment.
    Download PDF (2595K)
Technical Paper
  • Takeo Atsumi, Koshi Nishikawa, Raymond Munday
    2021 Volume 52 Issue 6 Pages 1329-1334
    Published: 2021
    Released: November 25, 2021
    JOURNAL FREE ACCESS
    In order to investigate vehicle driving behavior on soft soil, this study was focused on slip ratio, rolling resistance, traction force and drawbar pull. The wheel sensor for traction force measurement in actual driving condition was developed, and together with wheel speed, vehicle speed, and acceleration measurement, it enabled quantitative analysis of vehicle driving behavior on soft soil. This study explains the driving characteristics on soft soil and introduces efficient driving control method. And also explains tacit knowledge about driving on sand.
    Download PDF (547K)
Research Paper
  • Takashi Kondo, Kei Mochizuki, Kazuhiko Iso
    2021 Volume 52 Issue 6 Pages 1335-1342
    Published: 2021
    Released: November 25, 2021
    JOURNAL FREE ACCESS
    This paper investigates an "Exhilarating " engine sound quality for in-line four cylinder downsizing turbocharged engine. The downsizing turbo engine can realize quietness by lowering the rotational speed in combination with the Continuously Variable Transmission. On the other hand, noise and vibration caused by combustion tends to be a problem at low and middle rotational speed range due to high load operation. The relationship between the rumble sound generated at this engine operation condition and the target "Exhilarating " was clarified. And each part was promoted goal setting and countermeasures for achievement "Exhilarating " engine sound. Furthermore, we considered the difference between the sound quality "lightness " that was clarified in the previous research.
    Download PDF (1265K)
  • - Relationship between Blade Separation and Noise -
    Sho Kosaka, Kaori Seki, Masaharu Sakai, Hideaki Sato, Yoshiaki Nishiji ...
    2021 Volume 52 Issue 6 Pages 1343-1347
    Published: 2021
    Released: November 25, 2021
    JOURNAL FREE ACCESS
    The purpose of this research is to develop a technology to reduce the noise generated from the fan blades in order to reduce the air-conditioner blowing noise. Then, it aims at clarifying the noise generation mechanism of the cascade. The internal flow was observed in detail by conducting a PIV visualization test for a product equivalent cascade and a cascade with reduced interblade separation. In addition, by carrying out simultaneous measurement of noise, it was found that the size of noise could not be determined only by the size of separation.
    Download PDF (990K)
  • - Development of Modeling Method for Polymer Fracture Model -
    Kosuke Arita, Ryosuke Chikazawa, Kosho Kawahara
    2021 Volume 52 Issue 6 Pages 1348-1353
    Published: 2021
    Released: November 25, 2021
    JOURNAL FREE ACCESS
    This paper describes the development of analysis method for predicting a crack or a fracture in polymer parts. In recent years, demands to reduce the number of prototypes and vehicle scale tests by utilizing CAE analysis are increasing. Material model which considers the dependencies on stress-triaxiality and strain rate, was employed. Material tests with proposed test pieces were conducted to evaluate its validity and determine material model parameters for CAE analysis. Crash simulation with a component of vehicle parts was conducted to validate CAE modeling and analysis method. Crack initiation phenomena in polymer parts was validated with proposed analysis method.
    Download PDF (1020K)
  • Shuhei Kishino, Fukashi Sugasawa, Takashi Imaseki, Hiroshi Mouri
    2021 Volume 52 Issue 6 Pages 1354-1359
    Published: 2021
    Released: November 25, 2021
    JOURNAL FREE ACCESS
    In this paper, a path planning method is proposed that prepares to avoid the danger. Path planning is performed based on the acceleration for collision avoidance (ACA) that represents the severity of potential danger. ACA is the acceleration required to avoid collision with the object vehicle. This paper focuses on the driving scenario for overtaking. Vehicle driving scenario is simulated with proposed method. As a result, the ego vehicle's velocity and position are planned so that the vehicle can avoid object vehicle's lane change at any time. This path planning method and its simulation results are discussed.
    Download PDF (693K)
  • Takuma Yamaguchi, Hayato Kuroda, Hiroyuki Okuda, Tatsuya Suzuki, Kenta ...
    2021 Volume 52 Issue 6 Pages 1360-1367
    Published: 2021
    Released: November 25, 2021
    JOURNAL FREE ACCESS
    In the development of autonomous driving technology, simulation verification is an essential process, and pedestrian behavior is especially a key factor. However, the behavior and making-decision of humans are complex and indistinct. Therefore, the simulated behavior is assumed in a limited manner. To address this problem, we construct the pedestrian model of the crossing decision for a non-signalized intersection. The crossing decision is an interactive behavior that is influenced by other traffic participants. From this reason, we developed a multi-player interactive simulator, in which vehicles and a pedestrian can be controlled by the human. The model was constructed by the multi-class logistic regression to express human decision ambiguity. Finally, this model has validated whether the estimated decision was consistent with the ground truth data.
    Download PDF (1967K)
  • Shunya Seiya, Kento Ohtani, Alexander Carballo, Eijiro Takeuchi, Kazuy ...
    2021 Volume 52 Issue 6 Pages 1368-1374
    Published: 2021
    Released: November 25, 2021
    JOURNAL FREE ACCESS
    End-to-end driving refers to deep learning methods for generating control signals directly from external sensors. Previous methods use a direction vector towards the target to select and turn at intersections. However, the vector has a smaller dimension than the image, and thus it is ignored during training. In this study, we propose a learning method to emphasize that vector by using L2 regularization, which enables a robot to follow trajectories with branches. We validate the system's performance by conducting experiments using several driving scenarios. Our approach allowed an autonomous robot to successfully follow trajectories, including unknown outdoor trajectories.
    Download PDF (3796K)
feedback
Top