Correlation equalization-based automatic driving planning method and device for signal-free intersection

Abstract

The present disclosure provides a related equilibrium-based automatic driving planning method and device for a signal-free intersection. Vehicles in the signal-free intersection are modeled as non-cooperative games; according to the conflict of vehicles at the intersection, each conflict point is assigned an action from passing and yielding, forming an action combination corresponding to a strategy; the set of actionable actions in the modeling is obtained through the assigned permutations and combinations; according to the initial speed of the vehicles entering the intersection, the speed of each strategy is planned according to the principle of uniform passing and uniform deceleration yielding; according to the speed planning result, the vehicle motion trajectory of each strategy is determined, and the revenue is calculated; with the goal of maximizing revenue, a related equilibrium algorithm is used to solve the game strategy, and then the final strategy combination is selected, and each vehicle is informed of the recommended driving strategy. The application of the present application for vehicle decision-making at a signal-free intersection can reduce the occurrence of collisions and deadlocks and improve traffic efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic driving, in particular to a related-equalization-based automatic driving planning method and device for a signal-free intersection. BACKGROUND

[0002] The development of intelligent vehicles is of great significance to improve traffic efficiency, reduce congestion and improve traffic safety, and has made great progress. However, in some complex traffic environments, it is still severely limited, such as signal-free intersections. In the absence of traffic lights, the order of passage depends on the competition between drivers, which makes signal-free intersections prone to collision accidents, and accidents often cause traffic congestion, greatly reducing road efficiency and driving safety. The traditional automatic driving method adopts sequential planning, which is divided into four modules of perception, decision, planning and control. The decision planning module plays a decisive role in reducing congestion at intersections and other issues. The adjacent decision module and motion planning module are highly related in terms of function and result performance, and the influence of each other should be fully considered.

[0003] Due to the huge number of current vehicles, for a long time in the future, it will be a mixed traffic situation of manned and unmanned driving. Interaction with human drivers is crucial for autonomous vehicles, and even conservative driving behavior can cause traffic accidents due to misunderstanding by human drivers, so autonomous vehicles need to be more human, and their behavior and intentions should be more easily understood and accepted by human drivers. While ensuring safety, the decision of autonomous vehicles is also expected to help improve overall traffic efficiency without sacrificing their own interests.

[0004] Among the current decision-making methods of autonomous driving, the method based on machine learning is the most popular, including deep learning, reinforcement learning and transfer learning, etc. Various methods and model predictive control methods are combined to enable the decision-making model to adjust and optimize parameters according to lower-level motion characteristics. In order to learn a more reliable strategy, some scholars propose a reward scheme based on risk, which takes into account both utility and risk measurement. Compared with previous work that only uses collision and success as the only factor of the reward function, this risk-aware strategy can be more reliable and safer in challenging situations without being overly cautious. However, the machine learning-based method generally has the disadvantages of poor explainability, dependence on data acquisition, and long training time.

[0005] Furthermore, the challenges of human interaction in mixed traffic scenarios pose a challenge to autonomous driving decision-making. An intuitive solution is to adopt a rule-based approach, integrating the adaptive learning capabilities of machine learning with the interpretability of expert systems within a unified architecture to create a hybrid autonomous driving navigation system. However, rule-based interaction approaches rely heavily on pre-set rules, resulting in poor generalization performance.

[0006] A growing number of studies are using game theory to model the interactions between traffic participants. Most importantly, it provides insights into the rational decision-making of participants. The most common concept of a solution in game theory is the Nash equilibrium. Research typically models the interactions between agents as a best-response game, in which each agent negotiates to maximize its own utility and solves the game by finding a Nash equilibrium. Another common solution concept is the Stackelberg equilibrium. The ego vehicle is typically considered a leader, while obstacle vehicles are considered followers. Followers optimally respond based on the leader's behavior, while the leader, knowing this information, makes its own optimal response, forming a two-level optimization problem. Due to this characteristic of the Stackelberg game, it is more suitable for scenarios with a clear priority order, such as merging straight-moving vehicles with left-turning vehicles and lane changes. In most cases, Nash games perform better than Stackelberg games when surrounding agents are non-ideal. Level-k reasoning is also a game theory concept and a hierarchical decision-making concept that reflects the bounded rationality of agents. To improve overall road performance, some studies have adopted the grand coalition game approach to solve the decision-making problem of autonomous vehicles. However, cooperative games are not applicable in mixed traffic scenarios. Therefore, there is an urgent need for a game decision-making mechanism for autonomous driving vehicles that can increase overall benefits, ensure safety and improve traffic efficiency without violating individual rationality. Summary of the Invention

[0007] In view of this, the present invention proposes a method and device for autonomous driving planning and decision-making at unsignaled intersections based on correlated equilibrium, which effectively solves the decision-making problem of autonomous driving vehicles at unsignaled intersections. Compared with traditional methods, it has better explainability, reduces collisions and deadlocks, and improves overall road performance.

[0008] In order to solve the above technical problems, the present invention is implemented as follows.

[0009] A method for autonomous driving planning at an unsignaled intersection based on correlation equilibrium, comprising:

[0010] Step 1: Game modeling: The road participants in the unsignalized intersection are modeled as a non-cooperative game, consisting of a triple G = (N, A, u); where N = {1, ..., n} is the participant of the game, i.e., all vehicles; A = A1 × ... × A n , Ai is the actionable set of participant i; when the ego vehicle and other vehicles have future trajectories that intersect at the intersection, the vehicles have conflicts, and the actionable set is composed of actions assigned to each conflict point involved in participant i, and each conflict point can assign actions including passing and yielding; by assigning different actions to each conflict point and arranging combinations, a plurality of action combinations are obtained, which constitute the actionable set; each action combination is a strategy;

[0011] Step 2: According to the initial speed of the vehicle entering the intersection, the speed planning of each strategy passing through the intersection is performed: for each conflict point corresponding to a strategy, the passing action corresponds to the vehicle moving at a constant speed entering the intersection, and the yielding action corresponds to the vehicle uniformly decelerating to avoid conflict collision; if an action combination has more than 2 actions, the minimum acceleration in all actions is selected as the speed planning result of the entire intersection; according to the speed planning result of each strategy, the position of each vehicle at each time when passing through the intersection is calculated;

[0012] Step 3: According to the position, whether a collision occurs and the passing time are determined, and then the benefit of each vehicle under each strategy is calculated;

[0013] Step 4: Taking the maximization of benefit as the goal, a related equilibrium algorithm is used to solve the game strategy, to obtain the optimal probability distribution corresponding to the strategy combination of all vehicles, and according to the optimal probability distribution, the final strategy combination is selected, and each vehicle is recommended a driving strategy.

[0014] Preferably, in step 2, the determination of the acceleration a of the uniformly decelerating vehicle corresponding to the yielding action is: the acceleration a is designed so that when the other vehicle passes through the conflict point, the ego vehicle maintains the minimum safety distance from the other vehicle.

[0015] Preferably, the acceleration a is:

[0016]

[0017] where d ego is the distance from the ego vehicle to the conflict point, d safe is the minimum safety distance of the ego vehicle, v ego is the speed of the ego vehicle, and t1 is the time required for the other vehicle to arrive at the conflict point at a constant speed.

[0018] Preferably, in step 3, the benefit is calculated as the weighted sum of safety, passing efficiency and comfort.

[0019] Preferably, in the benefit calculation, the safety benefits u s of the two vehicles at the conflict point are obtained as follows:

[0020]

[0021] Among them, v1 and v2 are the initial velocities of the two vehicles entering the intersection; v max is the maximum speed allowed by the road; I(·) is an indicator function, which takes the value of 1 if (·) holds, otherwise it takes the value of 0; O c is the overlapping area of ​​the area occupied by the vehicle and the other vehicle during the entire passage process, O c >0 means that there is a collision between the vehicle and the other vehicle during the entire process of passing through the intersection; for any vehicle, a safety benefit is calculated for each vehicle that conflicts with it, and the sum of the safety benefits of all conflicting vehicles involved is the safety benefit of the vehicle.

[0022] Preferably, in the benefit calculation, the traffic efficiency benefit u e It is expressed using the time it takes for a vehicle to pass the last conflict point:

[0023] u e =-t

[0024] Where t is the time required for the ego vehicle to reach the conflict point.

[0025] Preferably, in the benefit calculation, the comfort benefit u c Use the change in acceleration and whether a collision occurs to express:

[0026]

[0027] Among them, |·| is the absolute value symbol; a is the acceleration of the vehicle; a min is the minimum acceleration allowed by the vehicle, i.e., the maximum deceleration; I(·) is an indicator function, which takes the value of 1 if (·) holds, otherwise it takes the value of 0; O c is the overlapping area of ​​the area occupied by the vehicle and other vehicles during the entire passage process, O c >0 means that the vehicle collides with other vehicles during the entire process of passing through the intersection, O c = 0, no collision occurs; when calculating O c When considering all vehicles that have potential conflicts with the vehicle, no matter which vehicle collides with it, it means O c >0.

[0028] Preferably, in step 4, the objective function of the correlation equilibrium algorithm aims to maximize the overall benefit while taking fairness into consideration, and the objective function is the social welfare term minus the social fairness term;

[0029] The social welfare term is: the total benefits of all vehicles under the strategy combination with corresponding probability;

[0030] The social fairness term is: calculate the square of the benefit difference between any two vehicles and then sum them.

[0031] Preferably, the determination of the collision of the two vehicles at the intersection is as follows:

[0032] Define the time interval in which the vehicle may cause a collision when passing through the conflict point; the start of the time interval is the time when the vehicle reaches a position that is half a car length plus half a car width away from the conflict point, and the end of the time interval is the time when the vehicle reaches a position that is more than half a car length plus half a car width away from the conflict point;

[0033] When the time intervals of the two vehicles intersect, it is determined that a collision occurs, otherwise the two vehicles pass through the intersection safely and no collision occurs.

[0034] The application also provides an automatic driving planning device for a signal-free intersection based on correlation equilibrium, which comprises a game modeling module, a speed planning module, a benefit calculation module, a correlation equilibrium solving module and a recommendation module.

[0035] The game modeling module is used to model the road participants in the signal-free intersection as a non-cooperative game; it is composed of a triple G=(N,A,u); wherein N={1,…,n} is the participant of the game, i.e. all vehicles; A=A1×…×An is the action set of the participant i; when the future trajectory of the ego vehicle intersects with that of other vehicles at the intersection, a conflict occurs, and the action set is composed of the actions assigned to each conflict point involved by the participant i, and each conflict point can assign actions including passing and yielding; by assigning different actions to each conflict point and arranging and combining them, a plurality of action combinations are obtained to form the action set; each action combination is a strategy; n i The speed planning module is used to plan the speed of each strategy for passing through the intersection according to the initial speed of the vehicle entering the intersection: for each conflict point corresponding to a strategy, the passing action corresponds to the vehicle moving at a constant speed with the speed of entering the intersection, and the yielding action corresponds to the vehicle uniformly decelerating to avoid collision; if an action combination has more than 2 actions, the minimum acceleration in all actions is selected as the speed planning result of the entire intersection; according to the speed planning result of each strategy, the position of each vehicle at each time when passing through the intersection is calculated.

[0036] The benefit calculation module is used to determine whether the vehicle may collide and the passing time according to the position, and then calculate the benefit of the vehicle under each strategy and send it to the correlation equilibrium solving module.

[0037] The benefit calculation module is used to determine whether the vehicle may collide and the passing time according to the position, and then calculate the benefit of the vehicle under each strategy and send it to the correlation equilibrium solving module.

[0038] ​The correlation equilibrium solving module is configured to maximize social welfare and social fairness, and to solve game strategies by using a correlation equilibrium algorithm to obtain an optimal probability distribution corresponding to a final strategy combination of all vehicles, and to select the final strategy combination according to the optimal probability distribution.

[0039] The recommendation module is configured to notify each vehicle of a driving strategy recommended by the final strategy combination.

[0040] Beneficial effects:

[0041] (1) The present application is based on the correlation equilibrium theory in game, thus having good explainability, and improving the overall performance of the road without violating the individual rationality, and performing excellently in avoiding collision and deadlock.

[0042] (2) In an optimal embodiment, when designing the objective function of the correlation equilibrium algorithm, the pursuit of fairness by human drivers is considered, and a fairness term is used to describe it, so that the decision result conforms to the psychology of human beings pursuing fairness.

[0043] (3) In an optimal embodiment, a specific acceleration value a of the uniform deceleration of the yielding strategy is given, and the parameters used are simple to obtain.

[0044] (4) In an optimal embodiment, the heterogeneity of human drivers is considered, and the driving style is considered by adding a driving style to the game payoff function, and the weight of each payoff factor is adjusted according to the driving style, so as to realize more human-like decision and personalized driving behavior.

[0045] (5) In an optimal embodiment, the efficiency payoff is described by the negative value -t of the passing time, so as to encourage vehicles to pass through the intersection faster, and since the vehicle trajectory corresponding to different strategies can be calculated, the value of t is easy to obtain, so that the present application is easy to implement. BRIEF DESCRIPTION OF DRAWINGS

[0046] Fig. 1(a) is a typical two-person signal-free intersection scene diagram;

[0047] Fig. 1(b) is a multi-person signal-free intersection scene diagram;

[0048] Figure 2 The flow chart of the signal-free intersection automatic driving planning method based on the correlation equilibrium of the present application;

[0049] Figure 3 The implementation framework of the overall method of the present application;

[0050] Figure 4 The simulation scene example diagram;

[0051] Figure 5A vehicle trajectory diagram obtained according to an algorithm;

[0052] Figure 6 A schematic diagram of an automatic driving planning device for a signal-free intersection based on correlation equalization according to the present application. DETAILED DESCRIPTION

[0053] The present application will be described in detail below with reference to the accompanying drawings and examples.

[0054] The present application provides a signal-free intersection automatic driving planning scheme based on correlation equalization, the basic idea of which is: for a signal-free intersection scenario, two actions are given for each conflict point, namely, yielding and passing, and the combination of actions of all conflict points is a strategy; according to the initial speed of the vehicle entering the intersection, the speed planning of each strategy through the intersection is carried out; the speed planning scheme of each strategy is determined; according to the speed planning result, the benefit of each vehicle under different strategies is calculated; then the correlation equalization algorithm is used to solve the game strategy, the optimal probability distribution corresponding to the strategy combination of all vehicles is obtained, the final strategy combination is selected according to the optimal probability distribution, and the recommended driving strategy for each vehicle is informed.

[0055] Referring to Figure 2 and Figure 3 , the specific implementation steps of the signal-free intersection automatic driving planning scheme of the present application are as follows:

[0056] Step 1, game modeling:

[0057] The road participants in the signal-free intersection are modeled as a non-cooperative game, which is composed of a three-tuple G=(N,A,u). Wherein,

[0058] N={1,…,n} is the participant or player of the game, i.e. all vehicles;

[0059] A=A1×…×A n , and A i is the action set of participant i∈N, in the intersection scenario, only deceleration can be avoided as much as possible to ensure the passing efficiency, and acceleration cannot be used to improve the efficiency, otherwise the optimization of efficiency will bring unlimited acceleration, resulting in dangerous driving behavior, therefore, in the intersection scenario, the action of the participant is two: passing or yielding.

[0060] A typical two-person scenario is shown in Fig. 1(a), which includes a target vehicle interacting with the ego vehicle, and the future trajectories of the two vehicles intersect, and the intersection point is called a conflict point. Fig. 1(b) includes the ego vehicle and two vehicles, and the future trajectories of the two vehicles intersect, and there are two conflict points. Whether there is a conflict point can be judged according to whether the future trajectories of the vehicles intersect.

[0061] In the case of two vehicles, as shown in Fig. 1(a), there is only one conflict point, and the possible actions are passing or yielding. In the case of multiple vehicles, as shown in Fig. 1(b), the possible actions are determined by the number of conflict points that the participants are involved in. At each conflict point, there are two possible actions, including passing and yielding. Each participant selects an action at each conflict point that he or she is involved in, and the combination of actions constitutes an action combination. By assigning different actions to each conflict point and performing permutation and combination, a plurality of action combinations are obtained, which constitute the set of possible actions. Considering that the participants adopt pure strategies, the possible actions are equivalent to strategies, and each action combination is equivalent to a strategy.

[0062] The above defines the actions of the participants and is equivalent to a strategy, s i represents the strategy of participant i, s i is selected from the strategy set S i of participant i. The Cartesian product of the strategy sets of all participants constitutes the set of mixed strategy combinations: S1×…×S n . It is defined that s -i =(s1,…,s i-1 ,s i+1 ,…,s n ) is a strategy combination without the strategy of participant i, so the strategy combination can be written as s=(s i ,s -i ). The strategy of each participant is selected from the strategy set S i .

[0063] u=(u1,…,u n ) and u i :A→R is the payoff of participant i∈N, involving safety, comfort, and passing efficiency, and has different emphases according to the different driving styles and needs of the participants.

[0064] Step 2, speed planning and position calculation:

[0065] In this step, according to the initial speed of the vehicle entering the intersection, the speed of each strategy passing through the intersection is planned according to the principle of uniform passing and uniform deceleration yielding.

[0066] The application designs two actions, passing and yielding. Taking the perspective of the ego vehicle as an example, the speed planning corresponding to the passing action is that the vehicle continues to start uniform motion at the speed of entering the intersection; the speed planning corresponding to the yielding action is that the vehicle uniformly decelerates to avoid collision. In the case of multiple people, the participants correspond to multiple conflict points, there are multiple action combinations, an action combination contains multiple actions, each action is speed planned in the aforementioned manner, and then the safest action (the smallest acceleration) is selected from the action combination as the speed planning result of the overall behavior of passing through the intersection. That is, when an action combination has 2 or more actions, the smallest acceleration corresponding to each action is selected as the speed planning result of the entire intersection.

[0067] Then, according to the speed planning result of each strategy, the position of each vehicle at each time when passing through the intersection is calculated. According to the position, whether a conflict occurs and the passing time can be calculated in step 3 to calculate the benefit.

[0068] In a preferred embodiment, a uniform deceleration expected acceleration a design scheme is provided as:

[0069]

[0070] where d ego is the distance from the ego vehicle to the conflict point, d safe is the minimum safety distance of the ego vehicle, v ego is the speed of the ego vehicle, and t1 is the time required for the opposite vehicle to travel at a uniform speed to reach the conflict point. The meaning of this acceleration is that when the opposite vehicle passes through the conflict point, the ego vehicle is just at the minimum safety distance from the opposite vehicle, so as to avoid collision. In practice, the yielding vehicle can also be designed to leave the intersection at the maximum acceleration after the opposite vehicle passes through the conflict point.

[0071] Step 3, according to the position, determine whether a conflict occurs and the passing time, and then calculate the benefit of the vehicle under each strategy.

[0072] First, the benefit function is designed: there are many factors that affect the driver's decision-making during driving at the intersection. First, safety must be ensured, no collision with other vehicles, second, the intersection should be passed as quickly as possible, and finally, in order to improve driving comfort, sudden acceleration and deceleration should be avoided as much as possible. The designed benefit function also starts from these three aspects.

[0073] (1) Safety benefit u s

[0074] Collision is not tolerated in the driving process, so the safety benefit depends on whether the vehicle collides with its interactive object. When a collision occurs, the degree of injury depends on the speed of both parties, so the safety benefit is designed as follows:

[0075]

[0076] where, u s is the safety benefit of the two vehicles that have conflict points, v1 and v2 are the initial speeds of the two vehicles that have conflict points entering the intersection; v max is the maximum speed allowed by the road, I(·) is an indicator function, which takes the value of 1 if (·) is true, and 0 otherwise. c is the overlap area of the two vehicles in the entire passing process, O c >0 means that there is a collision between the two vehicles throughout the entire process of passing through the intersection. For any vehicle, a safety benefit is calculated for each vehicle that conflicts with it, and the sum of the safety benefits of all the conflicting vehicles is the safety benefit of the vehicle.

[0077] It should be noted that when judging whether a collision occurs, the occupied area of the vehicle should be fully considered. Specifically, the present application considers a time interval, the start of the interval is the time when the vehicle reaches a position that is half a car length plus half a car width away from the conflict point, and the end of the interval is the time when the vehicle reaches a position that is more than half a car length plus half a car width away from the conflict point. The time interval depicts the entire time when the vehicle passing through the conflict point may cause a collision, and when the time intervals of the two vehicles have an intersection, it means that a collision occurs, otherwise it means that the two vehicles safely pass through the intersection without collision.

[0078] (2) Passing efficiency benefit u e

[0079] One of the core purposes of autonomous driving is to reduce people's commuting time, so passing efficiency is an important factor that needs to be considered in driving benefit design. From the perspective of the intersection scene, the passing efficiency can be described by the time when the vehicle passes through the conflict point, because after passing through the conflict point, the vehicle can move freely, accelerate or maintain a constant speed to pass through the intersection. From the perspective of the two parties involved, only the time before the conflict occurs is mutually affected, so this design is reasonable. Therefore, the passing efficiency benefit is designed using the time when the vehicle passes through the conflict point; if there are multiple conflict points, the time when the vehicle passes through the last conflict point is used to design the passing efficiency benefit as follows:

[0080] u e = -t

[0081] where t is the time for the ego vehicle to pass the conflict point. Under the go strategy, t is the time for the ego vehicle to reach the conflict point at constant speed; under the yield strategy, t is the time for the ego vehicle to reach the conflict point at constant deceleration. The go efficiency benefit is characterized by the negative value of the go time, thus encouraging the vehicle to pass the intersection faster. Since the trajectory of the vehicle under different strategies can be calculated, the value of t is easy to obtain.

[0082] (3) Comfort benefit u c

[0083] Comfort is also a commonly used index to characterize the driving benefit. Sudden braking, sudden stopping and frequent acceleration and deceleration during driving will bring bad experience to the passengers, so the change of acceleration is often used to characterize the comfort benefit. According to the speed planning method designed by the present application, the vehicle either travels at constant speed or adopts constant deceleration, so the change of acceleration can be represented by the size of the acceleration adopted. It is worth noting that when a collision occurs, the comfort brought is obviously the lowest, so it is given the minimum value. Therefore, the comfort benefit can be expressed by the change of acceleration and whether a collision occurs or not, and is expressed as follows:

[0084]

[0085] where |·| is the absolute value symbol. a is the acceleration of the ego vehicle, a = 0 corresponds to the go action, and the value of a corresponding to the yield action is calculated according to the formula above. a min is the minimum acceleration allowed by the vehicle, i.e. the maximum deceleration. I(·) is the indicator function, which takes the value 1 if (·) is true, and 0 otherwise; O c is the overlap area of the ego vehicle and the other vehicles in the entire go process. O c > 0 means that there is a collision between the ego vehicle and the other vehicles during the entire process of passing the intersection, O c = 0 means that there is no collision. When calculating O c , all vehicles with potential conflicts with the ego vehicle need to be considered, and whether a collision occurs with any vehicle means O c > 0.

[0086] In order to unify the dimensions, the above three benefits need to be normalized respectively. The final total benefit is the weighted sum of safety, go efficiency and comfort, and is expressed as follows:

[0087] u = ω s · u s + ω e · u e + ω c · u c

[0088] where ω s , ωe and ω c is a weighting factor. The total payoff function for each player is represented by the weighted sum of safety, efficiency, and comfort. Considering that human drivers have different driving styles—aggressive drivers tend to prioritize efficiency, while conservative drivers prioritize safety, even at the expense of efficiency—this invention represents these differences in driving style as different weighting coefficients in the payoff function.

[0089] Step 4: Correlated equilibrium solution: Use the correlated equilibrium algorithm to solve the game strategy, obtain the optimal probability distribution corresponding to the strategy combination of all vehicles, select the final strategy combination based on the optimal probability distribution, and inform each vehicle of the recommended driving strategy.

[0090] First, let's introduce the concept of correlated equilibrium. A distribution on S is a vector of non-negative real numbers, one for each strategy combination in S, summing to 1. A distribution x on S is a product if for each player p there is a distribution x p In S p So that all s=(s1,…,s n )∈S, there is A correlated equilibrium is a distribution x on S such that for all players p and all strategies i,j∈S p , the following is true:

[0091] Conditional on the pth part of the strategy combination extracted from x being i, the expected utility of player p executing i is not less than that of executing j, that is:

[0092]

[0093] Simply put, the expected return of executing the recommended strategy is no worse than any other strategy.

[0094] For convenience, use To represent the player p in strategy s=(s p ,s -p ) p (s p ,s -p ). Then, the correlated equilibrium condition can be expressed as the following linear program:

[0095]

[0096] According to the above conditions, any relevant equilibrium can be solved, so it is possible to optimize for certain goals. Since the present invention hopes that the decision result will improve the overall performance of the road, the following objective function is considered:

[0097]

[0098] where, The objective function represents the revenue maximization goal, which is the total revenue of all vehicles under the corresponding probability of the strategy combination. p represents the total revenue of player p. represents the strategy selected by player p in S p , and the strategies selected by other players in S -p . Player p and other players traverse the strategies in their own strategy set, calculate the revenue for each selection, and then add them together. represents the probability corresponding to each strategy selection.

[0099] Considering the psychological pursuit of fairness by humans, the invention further adds a fairness term, the social fairness term, to the objective function. Specifically, the revenue gap between players should be as small as possible, and for the convenience of calculation, the square of the difference between any two players is used to depict the fairness term, thereby forming a new objective function that is the social welfare term minus the social fairness term:

[0100]

[0101] where, is the social welfare term, which represents the revenue corresponding to the strategy combination that maximizes the total revenue of each vehicle; and i,j∈N,i≠j (u i -u j ) 2 is the social fairness term, which represents the square of the revenue difference between any two vehicles, and then sums them up.

[0102] Based on the above discussion, the solution process of the final game strategy is as follows, and the result obtained is a social welfare and social fairness related equilibrium (SWSFCE):

[0103]

[0104] Solving the above optimization problem can obtain the related equilibrium solution, i.e. a probability distribution of all strategy combinations, and then a fair related device selects the final strategy combination according to this probability distribution and informs each player of the recommended pure strategy, i.e. the recommended driving strategy for each vehicle.

[0105] Figure 4 The simulation environment for verifying the automatic driving planning and decision-making method of the invention at a signal-free intersection is shown. As shown in the figure, Figure 4 is a signal-free intersection scene, in which the ego vehicle enters the intersection from south to north, and a target vehicle beside it drives from west to east, and the future trajectories of the two vehicles conflict. The revenue matrix obtained by the algorithm of the invention is as follows:

[0106] Table 1 Profit Matrix

[0107] Yield Passage Yield 0.1115，0 0.6201，1 Passage 0.9663，0.4834 0.3663，0.4

[0108] The row player is the ego vehicle, and the column player is the target vehicle. The calculated correlated equilibrium strategy is (0, 0.6742, 0.3258, 0). After selection by the relevant devices, the (yield, pass) strategy is determined and sent to each vehicle. From the perspective of the ego vehicle, if it follows the agreement, its payoff is 0.6201. If it unilaterally violates the agreement, the target vehicle will maintain the pass, resulting in a (pass, pass) strategy. Its payoff is 0.3663, which is even lower. Therefore, the ego vehicle has no inclination to violate the agreement. From the perspective of the target vehicle, if it follows the agreement, its payoff is 1. If it unilaterally violates the agreement, the ego vehicle will maintain the yield, resulting in a (yield, yield) strategy. Its payoff is 0, which is even lower. Therefore, the target vehicle has no inclination to violate the agreement. This shows that the strategy outcome of the correlated equilibrium is stable, and neither player has an incentive to deviate from the agreement. Figure 5 The algorithm-derived trajectories of the two vehicles are shown. The ego vehicle is a rectangular box with horizontal lines inside, while the target vehicle is a blank rectangle. The ego vehicle yields, slowing down to allow the target vehicle to pass safely, and then accelerating through the intersection. During this process, the two vehicles do not collide.

[0109] Based on the above method, the present invention also provides an automatic driving planning device for an unsignaled intersection based on correlation equilibrium, such as Figure 6 As shown, it includes game modeling module, speed planning module, benefit calculation module, related equilibrium solution module and recommendation module.

[0110] The game modeling module is used to model the road participants in the unsignalized intersection as a non-cooperative game, which consists of a triple G = (N, A, u); where N = {1, ..., n} is the participant of the game, that is, all vehicles; A = A1 × ... × A n , A i is the set of possible actions for participant i. When the future trajectories of the vehicle and other vehicles intersect at an intersection and a conflict occurs, the possible action set consists of the actions assigned to each conflict point involved by participant i. Each conflict point can be assigned two actions: passing and yielding. By assigning different actions to each conflict point and performing permutations and combinations, multiple action combinations are obtained to form the possible action set. Each action combination is a strategy.

[0111] a speed planning module, configured to plan a speed for each strategy of passing the intersection according to an initial speed of the vehicle entering the intersection: for each conflict point corresponding to a strategy, the vehicle corresponding to an action moves at a constant speed with the speed of entering the intersection, and the vehicle corresponding to the action uniformly decelerates to avoid collision; if an action combination has more than two actions, the minimum acceleration in all actions is selected as the speed planning result of the entire intersection; and the position of each vehicle at each time when passing through the intersection is calculated according to the speed planning result of each strategy.

[0112] Here, in the yielding strategy, the acceleration a of the vehicle uniformly decelerating to avoid collision is calculated, and the design principle of the acceleration a is that the ego vehicle maintains the minimum safety distance from the opposite vehicle when the opposite vehicle passes through the conflict point, so as to avoid collision.

[0113] In a preferred embodiment, the calculation method of the acceleration a of the vehicle uniformly decelerating to avoid collision is as follows:

[0114]

[0115] wherein d ego is the distance from the ego vehicle to the conflict point, d safe is the minimum safety distance of the ego vehicle, v geo is the speed of the ego vehicle, and t1 is the time required for the opposite vehicle to uniformly travel to the conflict point.

[0116] a benefit calculation module, configured to determine whether the vehicles are likely to collide and the passing time according to the position, and then calculate the benefit of each vehicle under each strategy and send the benefit to a relevant equilibrium solving module.

[0117] In a preferred embodiment, the benefit is a weighted sum of safety, passing efficiency and comfort. The specific design of the safety benefit u s , the passing efficiency benefit u e and the comfort benefit u c is described in detail above, and will not be described here.

[0118] When calculating the above benefit factors, it is necessary to determine whether the two vehicles collide at the intersection, and the present application defines a time interval; the start of the time interval is the time when the vehicle reaches a position that is half a vehicle length plus half a vehicle width away from the conflict point, and the end of the time interval is the time when the vehicle reaches a position that is more than half a vehicle length plus half a vehicle width away from the conflict point; the time interval depicts all the time when the vehicle passes through the conflict point and may cause collision. When the time intervals of the two vehicles have intersection, it is determined that collision occurs, otherwise the two vehicles pass through the intersection safely without collision.

[0119] A correlation equilibrium solving module is configured to solve the game strategy by using a correlation equilibrium algorithm with the goal of maximizing the revenue, to obtain an optimal probability distribution corresponding to a strategy combination of all vehicles, and to select a final strategy combination according to the optimal probability distribution.

[0120] In a preferred embodiment, the goal of the correlation equilibrium algorithm is based on maximizing the revenue and taking fairness into account, and the objective function is designed as an objective function of a social welfare term minus a social fairness term. The social welfare term is the total revenue of all vehicles under the corresponding probability strategy combination, and the social fairness term is the square sum of the revenue difference between any two vehicles.

[0121] A recommendation module is configured to notify each vehicle of a driving strategy recommended by the final strategy combination according to the final strategy combination.

[0122] The specific embodiments above only describe the design principles of the present application, and the shapes and names of the components in the description can be different and are not limited. Therefore, those skilled in the art of the present application can modify or equivalently replace the technical solutions described in the foregoing embodiments; and these modifications and replacements do not deviate from the purpose and technical solutions of the present application, and should all belong to the protection scope of the present application.

Claims

1. A signal-free intersection automatic driving planning method based on correlation equalization, characterized in that, The method comprises the following steps: Step 1, game modeling: road participants in the intersection without signal are modeled as non-cooperative games, which are composed of a triple G = (N, A, u); where N = {1, …, n} is the participants of the game, i.e. all vehicles; A = A1x…xA n i is the action set of participant i; when the future trajectories of the ego vehicle and other vehicles intersect at the intersection, the vehicles conflict, and the action set is composed of actions assigned to each conflict point involved by participant i, and each conflict point can assign actions including passing and yielding; by assigning different actions to each conflict point and arranging combinations, a plurality of action combinations are obtained to form the action set; each action combination is a strategy;Step 2, game tree construction: the game tree is constructed by traversing the game tree from the root node to the leaf node, and the root node is the initial state of the game; the game tree is constructed by traversing the game tree from the root node to the leaf node, and the root node is the initial state of the game; Step 2: According to the initial speed of the vehicle entering the intersection, the speed planning of each strategy through the intersection is carried out: for each conflict point corresponding to a strategy, the action corresponding vehicle moves at a constant speed with the entering speed of the intersection, and the action corresponding vehicle uniformly decelerates to avoid collision; if an action combination has more than two actions, the minimum acceleration in all actions is selected as the speed planning result of the entire intersection; according to the speed planning result of each strategy, the position of each vehicle at each time when passing through the intersection is calculated; Step 3: According to the position, whether the vehicle is likely to collide and the passing time are determined, and then the benefit of the vehicle under each strategy is calculated; Step 4: A related equilibrium algorithm is used to solve the game strategy to obtain the optimal probability distribution corresponding to the strategy combination of all vehicles, and the final strategy combination is selected according to the optimal probability distribution, and each vehicle is informed of the recommended driving strategy; wherein the objective function of the related equilibrium algorithm aims to maximize the overall benefit and takes into account fairness, and the objective function is a social welfare term minus a social fairness term.

2. The method of claim 1, wherein, In step 2, the acceleration a of the action corresponding vehicle is determined as follows: the acceleration a is designed to make the ego vehicle maintain the minimum safety distance from the other vehicle when the other vehicle passes through the conflict point.

3. The method of claim 2, wherein, The acceleration a is: where d ego is the distance from the ego vehicle to the conflict point, d safe is the minimum safety distance of the ego vehicle, v ego is the speed of the ego vehicle, and t1is the time required for the opposing vehicle to reach the conflict point at a constant speed.

4. The method of claim 1, wherein, In step 3, the benefit is the weighted sum of safety, passing efficiency and comfort.

5. The method of claim 4, wherein, In the income calculation, the two vehicles with conflict points have their respective safety income u s The acquisition method is: where v1 and v2 are the initial speeds of the two vehicles that collide at the intersection; v max is the maximum speed allowed on the road; I(·) is the indicator function, which takes the value 1 if (·) is true, and 0 otherwise; O c is the overlap area of the ego vehicle and the other vehicle in the whole process of passing through the intersection, O c > 0 means that there is a collision between the ego vehicle and the other vehicle in the whole process of passing through the intersection; for any vehicle, a safety benefit is calculated for each conflicting vehicle, and the sum of the safety benefits of all the involved conflicting vehicles is the safety benefit of the vehicle.

6. The method of claim 5, wherein, The passage efficiency benefit u e Expressed by the time of the vehicle passing the last conflict point: u e = -t Where t is the time required for the ego vehicle to reach the conflict point.

7. The method of claim 5, wherein, In the benefit calculation, the comfort benefit u c Using the change in acceleration and whether a collision occurs to express: where |·| is the absolute value symbol; a is the acceleration of the ego vehicle; a min is the minimum acceleration allowed for the vehicle, i.e., the maximum deceleration; I(·) is the indicator function, which takes the value 1 if (·) is true and 0 otherwise; O c is the overlap area of the ego vehicle and the other vehicle in the entire passage, O c > 0 means that there is a collision between the ego vehicle and the other vehicle in the entire passage, O c = 0 means that there is no collision. In calculating O c , all vehicles that have a potential conflict with the ego vehicle are considered, and O c > 0 means that there is a collision with any of them.

8. The method of claim 1, wherein, In step 4: The social welfare term is: the overall benefit of all vehicles under the corresponding probability strategy combination; The social fairness term is: the square sum of the benefit difference between any two vehicles.

9. The method of claim 1, wherein, The determination method of collision between two vehicles at the intersection is as follows: Define the time interval that may cause collision when the vehicle passes through the conflict point; the start of the time interval is the time when the vehicle reaches the position of half a car length plus half a car width away from the conflict point, and the end of the time interval is the time when the vehicle reaches the position of more than half a car length plus half a car width away from the conflict point; When the time intervals of the two vehicles have intersection, it is determined that collision occurs, otherwise the two vehicles pass through the intersection safely without collision.

10. A signal-free intersection automatic driving planning device based on correlation equalization, characterized by, The method comprises a game modeling module, a speed planning module, a benefit calculation module, a related equilibrium solving module and a recommendation module; The game modeling module is configured to model the road participants in the unsignalized intersection as a non-cooperative game; the game is composed of a triple G=(N, A, u); wherein N={1,...,n} is the participants of the game, i.e., all vehicles; A=A1x...xA n , A i is the set of actions available to participant i; when the future trajectories of the ego vehicle and other vehicles intersect at the intersection, the vehicles collide, and the set of actions available to participant i consists of actions assigned to each conflict point involved by the participant i, and each conflict point can assign actions including passing and yielding; a plurality of action combinations are obtained by assigning different actions to each conflict point and performing permutation and combination, to form the set of actions available; each action combination is a strategy; The speed planning module is configured to plan the speed of each strategy through the intersection according to the initial speed of the vehicle entering the intersection: for each conflict point corresponding to a strategy, the action corresponding vehicle moves at a constant speed with the entering speed of the intersection, and the action corresponding vehicle uniformly decelerates to avoid collision; if an action combination has more than two actions, the minimum acceleration in all actions is selected as the speed planning result of the entire intersection; according to the speed planning result of each strategy, the position of each vehicle at each time when passing through the intersection is calculated; The benefit calculation module is configured to determine whether the vehicle is likely to collide and the passing time according to the position, and then calculate the benefit of the vehicle under each strategy and send it to the related equilibrium solving module; The correlation equilibrium solving module is configured to solve the game strategy by using a correlation equilibrium algorithm, to obtain an optimal probability distribution corresponding to a final strategy combination of all vehicles according to the optimal probability distribution, and to select the final strategy combination; wherein, the objective function of the correlation equilibrium algorithm aims to maximize the overall benefit and takes into account fairness, and the objective function is a social welfare item minus a social fairness item. The recommendation module is configured to notify each vehicle of a driving strategy recommended by the final strategy combination for the vehicle according to the final strategy combination.

Images