WO2021148113A1 - Computing system and method for training a traffic agent in a simulation environment - Google Patents
Computing system and method for training a traffic agent in a simulation environment Download PDFInfo
- Publication number
- WO2021148113A1 WO2021148113A1 PCT/EP2020/051486 EP2020051486W WO2021148113A1 WO 2021148113 A1 WO2021148113 A1 WO 2021148113A1 EP 2020051486 W EP2020051486 W EP 2020051486W WO 2021148113 A1 WO2021148113 A1 WO 2021148113A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- trajectories
- vehicle
- trajectory
- term
- long
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
- G09B—EDUCATIONAL OR DEMONSTRATION APPLIANCES; APPLIANCES FOR TEACHING, OR COMMUNICATING WITH, THE BLIND, DEAF OR MUTE; MODELS; PLANETARIA; GLOBES; MAPS; DIAGRAMS
- G09B9/00—Simulators for teaching or training purposes
- G09B9/02—Simulators for teaching or training purposes for teaching control of vehicles or other craft
-
- G—PHYSICS
- G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
- G09B—EDUCATIONAL OR DEMONSTRATION APPLIANCES; APPLIANCES FOR TEACHING, OR COMMUNICATING WITH, THE BLIND, DEAF OR MUTE; MODELS; PLANETARIA; GLOBES; MAPS; DIAGRAMS
- G09B9/00—Simulators for teaching or training purposes
- G09B9/006—Simulators for teaching or training purposes for locating or ranging of objects
Definitions
- Perception/Map generally relates to the input about the environment that is available to other components.
- Traffic Rules generally relates to any component that provides legal restrictions to high-level decisions.
- Traffic-Free Reference Line generally relates to planning an “optimal" reference-line ignoring other traffic participants.
- Behavior Planning generally relates to planning a behavior plan, that is when exactly to conduct actions, such as lane changes, incorporating other participants.
- PCB Prediction- and Cost-function based
- a third aspect of the invention relates to a computing system for simulating a road driving environment in a driving situation including a complex driving situation for one or more vehicles.
- the computing system comprises or consists of one or more processors, a memory device coupled to the one or more processors, and a traffic agent using a neural network model for decision making in a driving situation including a complex driving situation stored in the memory device and configured to be executed by the one or more processors.
- the traffic agent is trained according to the inventive computer- implemented training to select as an action a candidate trajectory x c to be performed by the traffic agent, wherein the preference value y c respectively assigned to the selected candidate trajectory x c exceeds a predetermined threshold value.
- the expression “driving data per time frame t over a plurality of time frames t, for one or more road vehicles respectively driven by a human in a realistic situation on a road’ means that status data of the one or more road vehicles are provided per respective given time frame ti, i.e. the respective time frames (time stamps).
- the index / in ti represents the number of the respective time frame (time stamp), i.e. an integer of 1 , 2, 3 or more.
- the plurality of given time frames ti may have different or the same, preferably the same intervals, wherein the time interval between the plurality of the given time frames ti is generally as short as suitable, e.g. 0.04 seconds.
- step a) of the first inventive aspect driving data per time frame t over a plurality of time frames ti for one or more road vehicles respectively driven by a human in a realistic situation on a road is provided and for at least part of the road vehicles as ego vehicles a respective long-term ground truth trajectory x is generated.
- any suitable preference threshold value for selection of the candidate trajectory x c can be used.
- the predetermined threshold value for the preference value y c of the selected candidate trajectory x c is set to 50 % or more, alternatively 60 % or more, alternatively 80 % or more, alternatively 90 % or more in relation to the ground truth preference value y.
- the assigned preference value y c of the selected candidate trajectory x c exhibits 50 % or more, alternatively 60 % or more, alternatively 80 % or more, alternatively 90 % or more of the respective ground truth preference value y, which is set to 100%.
- relative bearing of a vehicle in the six-vehicle neighborhood represents in the context of the present invention the relative orientation of the respective vehicle with respect to the ego vehicle, in particular represented by the relative angle Q formed in the position of the ego vehicle in relation to the straight between the ego vehicle position and the position of the respective vehicle of the six-vehicle neighborhood in the global x- / y- axes and the respective x- axis.
- An additional or alternative preferred embodiment of the present invention relates to an embodiment, wherein 5 or more, alternatively 10 or more, alternatively 20 or more, alternatively 30 or more, alternatively 40 or more, alternatively 50 or more random alternative trajectories x; are generated uniformly sampled in bounds for the distance parameter space Q, and wherein 5 or more, alternatively 10 or more, alternatively 20 or more, alternatively 30 or more, alternatively 40 or more, alternatively 50 or more random alternative trajectories x; are generated sampled from a normal distribution around the distance parameter space ⁇ with the same constraints.
- a third aspect of the invention relates to a computing system for simulating a road driving environment in a driving situation including a complex driving situation for one or more vehicles comprising or consisting of one or more processors, a memory device coupled to the one or more processors, and a traffic agent using a neural network model for decision making in a driving situation including a complex driving situation stored in the memory device and configured to be executed by the one or more processors, characterized in that the traffic agent is trained according to the computer-implemented training method of the first inventive aspect to select as an action a candidate trajectory Xc, which exceeds the predetermined threshold value for the preference value y c to be performed by the traffic agent in the simulation environment.
- Figure 1a shows a graphical representation of an architecture embodiment 1 , also called “driving stack”, of a training computing system according to the present invention
- Figure 1b shows a graphical representation of an architecture embodiment 1’ of a simulation computing system according to the present invention.
- the inventive driving stacks are configured to make feasible human-like (naturalistic) decisions in all simulated driving situations including complex driving situations.
- the training architecture 1 displayed in Figure 1a) is configured to comprise the functions of module A and module B as described above with respect to the inventive training computer system in one module 11. Accordingly the combined module 11 is configured to perform the functions of module A, e.g., is configured to input and process the naturalistic driving data and to generate the ground truth trajectories x.
- the simulation architecture T displayed in Figure 1b) is configured to also comprise the functions of module A and module B as described above with respect to the inventive simulation computer system in one module 1 T.
- the combined module 1 T is configured to perform the functions of module A, e.g., is configured to simulated driving data of simulated vehicles in a simulation environment. This means that during deployment in a simulation environment module A does not use the naturalistic driving data and does not use the ground truth trajectories x.
- the combined module 1 T is configured to perform the functions of module B, e.g., is configured to generate perception values p of the simulated driving data per given time frame Module 12’ (module C), which is according to Figure 1 b) also called “Gen.
- TP has been trained with the inventive training method and is configured to generate one or more feasible alternative long-term trajectories x; per given time frames ft respectively assigned with a preference value yi Module 13’ (module D), which is according to Figure 1b) also called “FILDM”, represents a computer model for high level decision making in a driving situation including a complex driving situation and has been trained with the inventive training method to select as an action a candidate trajectory x c , which exceeds the predetermined threshold for the preference value y c to be performed by the traffic agent.
- module D which is according to Figure 1b
- FILDM represents a computer model for high level decision making in a driving situation including a complex driving situation and has been trained with the inventive training method to select as an action a candidate trajectory x c , which exceeds the predetermined threshold for the preference value y c to be performed by the traffic agent.
- FIG. 1b does not show that the modules A and B can alternatively be provided as separate modules.
- the inventors used the first four hours of driving data of the DataFromSky (DFS) data set (purchased from RCE systems s.r.o., Czech Republic), which forms a naturalistic driving data set for analysis.
- the naturalistic data comprised coordinate and velocity information of actual human driven trajectories on a small patch of about 500 m of the A9 Highway in Germany.
- the data in particular comprised per time frame a unique car identifier, global coordinates in lateral and longitudinal direction, longitudinal velocity and longitudinal acceleration.
- the inventors did not consider vehicles that are visible in less than 200 time frames ti or are not passenger cars (but e.g. trucks) as ego vehicles with respect to the experimental analysis below. From the first three hours of driving data, the inventors randomly sampled 72% as training data, 20% for testing and 8% for validation. For each resulting frame of data the inventors considered each car as ego vehicle, extracted its perception p, as described in the detailed description of the invention above and set out herein below and generated alternative long-term trajectories x; as described in the detailed description of the invention above and set out herein below. Accordingly, the inventors considered that the behavior of the driver of an ego vehicle is to a large extent a function of the state of the vehicle and that of the closest surrounding vehicles.
- DFS naturalistic driving data set
- This data set contains for a number of driving situations a ground truth trajectory x.
- Naturalistic trajectories may exhibit various (low-level) features that a model could over fit on. In this case, care should be taken to preferably not expose those (low-level) features during model training.
- the vehicles future ground truth trajectory x is accordingly extracted as an accumulation of data of the ego vehicle over time.
- the ground truth trajectory x is then transformed into Frenet Frame coordinates (for details see, Moritz Werling, Julius Ziegler, Soren Kammel, and Sebastian
- Table 2 gives an overview of the considered computational models together with the class balance during training.
- TP execution of module 14’ should not drastically interfere with the decision made by module 13’ (HLDM).
- HLDM decision made by module 13’
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Educational Administration (AREA)
- Educational Technology (AREA)
- General Physics & Mathematics (AREA)
- Aviation & Aerospace Engineering (AREA)
- Traffic Control Systems (AREA)
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/EP2020/051486 WO2021148113A1 (en) | 2020-01-22 | 2020-01-22 | Computing system and method for training a traffic agent in a simulation environment |
DE112020006317.8T DE112020006317T5 (de) | 2020-01-22 | 2020-01-22 | Computersystem und verfahren zum trainieren eines verkehrsagenten in einer simulationsumgebung |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/EP2020/051486 WO2021148113A1 (en) | 2020-01-22 | 2020-01-22 | Computing system and method for training a traffic agent in a simulation environment |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2021148113A1 true WO2021148113A1 (en) | 2021-07-29 |
Family
ID=69190785
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/EP2020/051486 WO2021148113A1 (en) | 2020-01-22 | 2020-01-22 | Computing system and method for training a traffic agent in a simulation environment |
Country Status (2)
Country | Link |
---|---|
DE (1) | DE112020006317T5 (de) |
WO (1) | WO2021148113A1 (de) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114820971A (zh) * | 2022-05-05 | 2022-07-29 | 吉林大学 | 一种描述复杂驾驶环境信息的图形化表达方法 |
-
2020
- 2020-01-22 DE DE112020006317.8T patent/DE112020006317T5/de active Pending
- 2020-01-22 WO PCT/EP2020/051486 patent/WO2021148113A1/en active Application Filing
Non-Patent Citations (8)
Title |
---|
ANONYMOUS: "Continental Invests in Virtual Development for Automated Driving and Collaborates with AAI", 9 January 2019 (2019-01-09), pages 1 - 2, XP055718324, Retrieved from the Internet <URL:https://www.continental.com/en/press/press-releases/virtual-development-157198> [retrieved on 20200727] * |
HANWOOL WOOYONGHOON JIYUSUKE TAMURAYASUHIDE KURODATAKASHI SUGANOYASUNORI YAMAMOTOATSUSHI YAMASHITAHAJIME ASAMA: "Trajectory prediction of surrounding vehicles considering individual driving characteristics", INTERNATIONAL JOURNAL OF AUTOMOTIVE ENGINEERING, vol. 9, no. 4, 2018, pages 282 - 288 |
ISABEL METZ: ""Create a Replica of the World" Using HD Maps for the Virtual Testing of HAD", 13 June 2019 (2019-06-13), 1st NDS Public Conference, pages 1 - 13, XP055718340, Retrieved from the Internet <URL:https://nds-association.org/wp-content/uploads/2019/06/NDS-Conference-2019__Create_a_Replica_of_the_World_AAI.pdf> [retrieved on 20200727] * |
JUNQING WEIJARROD M. SNIDERTIANYU GUJOHN DOLANBAKHTIAR LITKOUHI, A BEHAVIORAL PLANNING FRAMEWORK FOR AUTONOMOUS DRIVING, June 2014 (2014-06-01), pages 458 - 464 |
MORITZ WERLINGJULIUS ZIEGLERSOREN KAMMELSEBASTIAN THRUN, OPTIMAL TRAJECTORY GENERATION FOR DYNAMIC STREET SCENARIOS IN A FRENET FRAME, June 2010 (2010-06-01), pages 987 - 993 |
NACHIKET DEOMOHAN M. TRIVEDI., MULTI-MODAL TRAJECTORY PREDICTION OF SURROUNDING VEHICLES WITH MANEUVER BASED ISTMS. CORR, ABS/1805.05499, 2018 |
S. YOOND. KUM.: "The multilayer perceptron approach to lateral motion prediction of surrounding vehicles for autonomous vehicles", IN 2016 IEEE INTELLIGENT VEHICLES SYMPOSIUM (IV, June 2016 (2016-06-01), pages 1307 - 1312, XP032939137, DOI: 10.1109/IVS.2016.7535559 |
YEPING HUWEI ZHANMASAYOSHI TOMIZUKA, PROBABILISTIC PREDICTION OF VEHICLE SEMANTIC INTENTION AND MOTION. CORR, ABS/1804.03629, 2018 |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114820971A (zh) * | 2022-05-05 | 2022-07-29 | 吉林大学 | 一种描述复杂驾驶环境信息的图形化表达方法 |
CN114820971B (zh) * | 2022-05-05 | 2023-06-09 | 吉林大学 | 一种描述复杂驾驶环境信息的图形化表达方法 |
Also Published As
Publication number | Publication date |
---|---|
DE112020006317T5 (de) | 2022-11-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11900797B2 (en) | Autonomous vehicle planning | |
CN106874597B (zh) | 一种应用于自动驾驶车辆的高速公路超车行为决策方法 | |
US20230124864A1 (en) | Graph Representation Querying of Machine Learning Models for Traffic or Safety Rules | |
Bahram et al. | A game-theoretic approach to replanning-aware interactive scene prediction and planning | |
EP3977226A1 (de) | Vorrichtung, verfahren und artikel zur bewegungsplanung in einer umgebung mit dynamischen objekten | |
Yoon et al. | Predictive cruise control using radial basis function network-based vehicle motion prediction and chance constrained model predictive control | |
Ghodsi et al. | Generating and characterizing scenarios for safety testing of autonomous vehicles | |
CN113581182B (zh) | 基于强化学习的自动驾驶车辆换道轨迹规划方法及*** | |
Tehrani et al. | General behavior and motion model for automated lane change | |
CN112046484A (zh) | 一种基于q学习的车辆变道超车路径规划方法 | |
JP7520444B2 (ja) | 乗り物に基づくデータ処理方法、データ処理装置、コンピュータ機器、及びコンピュータプログラム | |
CN114061581A (zh) | 通过相互重要性对自动驾驶车辆附近的智能体排名 | |
CN114987498B (zh) | 自动驾驶车辆的拟人化轨迹规划方法、装置、车辆及介质 | |
Hart et al. | Formulation and validation of a car-following model based on deep reinforcement learning | |
Guo et al. | Toward human-like behavior generation in urban environment based on Markov decision process with hybrid potential maps | |
Lodhi et al. | Autonomous vehicular overtaking maneuver: A survey and taxonomy | |
WO2021148113A1 (en) | Computing system and method for training a traffic agent in a simulation environment | |
Dubey et al. | Autonomous braking and throttle system: A deep reinforcement learning approach for naturalistic driving | |
Kaur et al. | Scenario-based simulation of intelligent driving functions using neural networks | |
Sukthankar et al. | Evolving an intelligent vehicle for tactical reasoning in traffic | |
WO2023187121A1 (en) | Simulation-based testing for robotic systems | |
Mohammad et al. | Modeling Gap Acceptance in Overtaking: A Cognitive Process Approach | |
Chen et al. | Platoon separation strategy optimization method based on deep cognition of a driver’s behavior at signalized intersections | |
JP2023531927A (ja) | 運転意思決定方法および運転意思決定装置ならびにチップ | |
WO2021160273A1 (en) | Computing system and method using end-to-end modeling for a simulated traffic agent in a simulation environment |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 20701734 Country of ref document: EP Kind code of ref document: A1 |
|
DPE1 | Request for preliminary examination filed after expiration of 19th month from priority date (pct application filed from 20040101) | ||
122 | Ep: pct application non-entry in european phase |
Ref document number: 20701734 Country of ref document: EP Kind code of ref document: A1 |