Search

CN-122009237-A - Pedestrian implicit cooperative game method and system for unprotected intersection traffic

CN122009237ACN 122009237 ACN122009237 ACN 122009237ACN-122009237-A

Abstract

The invention provides a pedestrian implicit cooperative game method and system for unprotected intersection traffic, the method comprises the steps of sensing unprotected intersection environment, obtaining pedestrian state information and vehicle state information, modeling interaction processes of vehicles and pedestrians into a part of observable Markov decision processes based on state space, action space, observation space and a reward function, wherein the state space comprises vehicle state information and pedestrian state information, the action space comprises vehicle planning tracks, the reward function comprises a reward calculation for cooperation degree of vehicles and pedestrians, generating a plurality of candidate strategies with intention expression, predicting pedestrian reaction by using a pedestrian reaction prediction model for the candidate strategies, selecting an optimal strategy based on the reward function, executing the selected strategy, and performing rolling optimization according to real pedestrian reaction. The method enables the intelligent driving vehicle to actively and safely carry out implicit negotiation with pedestrians, and achieves efficient, smooth and safe unprotected intersection passing.

Inventors

  • LI ZHONGWEI

Assignees

  • 武汉江夏楚能汽车技术研发有限公司

Dates

Publication Date
20260512
Application Date
20260127

Claims (10)

  1. 1. The hidden collaborative game method for pedestrians passing through unprotected intersections is characterized by comprising the following steps of: S1, sensing an unprotected intersection environment through a vehicle-mounted sensor, and acquiring pedestrian state information and vehicle-mounted state information; S2, modeling an interaction process of a vehicle and a pedestrian as a partially observable Markov decision process based on a state space, an action space, an observation space and a reward function, wherein the state space comprises the vehicle state information and the pedestrian state information, the action space comprises a planning track of the vehicle, the observation space is a local observation space of the action of the pedestrian on the vehicle, and the reward function comprises a reward calculation for the cooperation degree of the vehicle and the pedestrian; S3, generating a plurality of candidate strategies with intention expression based on the part of observable Markov decision process, wherein each candidate strategy corresponds to a vehicle motion track and is endowed with interaction semantics; s4, for the candidate strategies, predicting pedestrian reaction by utilizing a pedestrian reaction prediction model, and selecting an optimal strategy based on the reward function; and S5, executing the selected optimal strategy, and performing rolling optimization according to the real response of the pedestrians.
  2. 2. The method according to claim 1, wherein the sensor in step S1 comprises a lidar, a camera or a millimeter wave radar; the pedestrian status information includes position, speed, acceleration, walking direction, and head orientation; The vehicle status information includes position, speed, acceleration, and jerk.
  3. 3. The method of claim 1, wherein the reward function in step S2 is a multi-objective function including at least a security reward, a traffic efficiency reward, and a synergy reward.
  4. 4. The method of claim 1, wherein the candidate policies in step S3 include a look-ahead policy, a courtesy policy, or a cooperative passing policy.
  5. 5. The method according to claim 1, wherein step S4 specifically includes, for each of the candidate strategies, predicting pedestrian reaction using a pedestrian reaction prediction model, outputting a probability distribution of a future trajectory of the pedestrian based on historical interaction semantic data, and selecting a strategy for which a cumulative prize is desired to be maximized as an optimal execution strategy.
  6. 6. The method of claim 5, wherein the pedestrian reaction prediction model is a deep learning network, the pedestrian behavior pattern is learned by training data, and a probability distribution including stopping, accelerating, or walking normally is output.
  7. 7. The method of claim 1, wherein the rolling optimization in step S5 includes continuously sensing pedestrian state changes and repeating steps S1 through S4 at regular time intervals, re-planning for closed loop control.
  8. 8. A pedestrian implicit collaborative gaming system for unprotected intersection traffic for implementing the method of any one of claims 1 to 7, comprising: the sensing module is configured to sense the unprotected intersection environment, the pedestrian state and the vehicle-self state through the sensor; the cooperative game decision module comprises a modeling unit, a strategy generation unit, a pedestrian reaction predictor and a strategy evaluation and selection unit, and is configured to execute modeling, candidate strategy generation, pedestrian reaction prediction and strategy selection; the track planning and control module is configured to convert the optimal strategy track into an executable path and speed and send the executable path and speed to the vehicle drive-by-wire system; and the rolling optimization module is configured to execute rolling optimization so as to perform closed-loop control.
  9. 9. An electronic device comprising at least one processor and a memory communicatively coupled to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the at least one processor implementing the pedestrian implicit collaborative gaming method for unprotected intersection traffic of any one of claims 1-7 by executing the instructions stored by the memory.
  10. 10. A computer readable storage medium having stored therein at least one instruction or at least one program loaded and executed by a processor to implement the pedestrian implicit collaborative gaming method for unprotected intersection traffic of any one of claims 1-7.

Description

Pedestrian implicit cooperative game method and system for unprotected intersection traffic Technical Field The invention relates to the technical field of automatic driving, in particular to a pedestrian implicit collaborative game method and system for unprotected intersection traffic. Background In the technical field of automatic driving, unprotected intersection passing is a key and complex application scene, and is characterized in that vehicles and pedestrians need to negotiate road rights through dynamic interaction under the condition of lacking clear traffic rules (such as traffic lights) constraint. Currently, when an automatic driving system processes unprotected intersection traffic, the automatic driving system mainly depends on two types of technical schemes, namely a passive decision method based on a prediction-response mode and an intention transmission mode based on explicit communication. Although the schemes realize basic functions to a certain extent, the schemes can not effectively simulate an inherent implicit cooperative mechanism in human driving, so that the performance and the reliability of the schemes in practical application are limited, and specifically: First, the predictive-response mode suffers from excessive conservation and passive drawbacks. The existing method generally regards pedestrian behaviors as independent prediction objects, adopts step-by-step processing, and predicts future tracks of pedestrians through sensor data and then performs vehicle path planning based on prediction results. This approach is passive in nature, and once the prediction accuracy is inadequate (e.g., due to multi-modal uncertainty of pedestrian behavior), the vehicle is prone to decision hesitation, manifested as sudden braking or long-term stalling, resulting in intersection "stuck" phenomena. The traffic efficiency is reduced, the traffic jam at the rear side can be caused, and the core requirements of the automatic driving system on the fluency and the safety are violated. The fundamental problem is that the vehicle and the pedestrian cannot be regarded as an interactive whole, but respond based on isolated predictions, and the active guiding capability is lacked. Second, explicit communication methods have practical and general limitations. The vehicle intention is transmitted to pedestrians through the outside of the vehicle (such as a display screen and light signals) so as to realize explicit communication, however, the method relies on the correct interpretation and trust of the pedestrians on the outside signals and faces multiple challenges in practical application, namely firstly, the pedestrians can have deviation on the understanding of non-standardized signals, and information confusion is easy to cause in a scene of multiple traffic participants, and secondly, the technology is not popular yet, and involves legal compliance and standardization problems and is difficult to deploy on a large scale. Thus, explicit communication cannot reliably replace natural implicit interactions in human driving, limiting its applicability in complex urban environments. Third, the prior art lacks a true interactive collaboration mechanism. Neither the prediction-response mode nor the explicit communication models the vehicle and the pedestrian as a collaborative decision unit, and these methods ignore the implicit collaborative process of delivering intent through vehicle motion (e.g., creep, acceleration change) in human driving, which is critical to achieving efficient, smooth traffic. The existing scheme often simplifies interaction into unidirectional or inefficient instruction transmission, and fails to simulate the mercy negotiation among human drivers, so that the behavior of the vehicle is stiff and the predictability is poor, and the safety and the comfort of the whole traffic flow are affected. In summary, the prior art is mainly limited by the core problems of passivity, communication dependence, and co-operation deficiency when processing unprotected intersection traffic, and a technical scheme capable of simulating human implicit interaction logic is urgently needed in the art, so that an intelligent driving vehicle can make a co-decision with pedestrians in an active and safe manner. Disclosure of Invention The invention aims to provide a pedestrian implicit cooperative game method and system for passing through unprotected intersections, and the technical problem that intelligent driving vehicles in the prior art are passive and low in efficiency in the unprotected intersections is solved. The technical scheme for solving the technical problems is as follows: in a first aspect, the present invention provides a method for implicit collaborative gaming for pedestrians passing through unprotected intersections, comprising the steps of: S1, sensing an unprotected intersection environment through a vehicle-mounted sensor, and acquiring pedestrian state information and vehicle-mount