Automatic driving right-of-way optimization lane changing method based on RSS (Received Signal Strength)

Abstract

The invention provides an RSS (Received Signal Strength)-based automatic driving right-of-way optimization lane changing method. The method comprises the following steps: establishing a responsibility sensitive safety model RSS model, wherein the RSS model comprises a longitudinal minimum safety distance set by using surrounding environment information and vehicle state information of an automatic driving vehicle (AV), and a driving distance safety constraint condition that the longitudinal distance from a surrounding vehicle to the AV is smaller than the minimum safety distance; constructing a decision making model, and training the decision making model by using a set global reward function and a vehicle lane changing performance evaluation index to obtain a trained decision making model; and designing a road right priority strategy by using a driving distance safety constraint condition through the trained decision making model, and carrying out AV road right priority lane changing decision by using the road right priority strategy according to the surrounding environment information and the road right lane changing decision evaluation feedback information. According to the method, the efficiency of the vehicle can be dynamically evaluated and optimized, lane changing safety is ensured, safety and efficiency are balanced, and road right optimization requirements of different vehicles are met.

Description

Technical Field

[0001] The present invention relates to the technical field of AV (Autonomous vehicles), and in particular to an automatic driving right-of-way optimization lane changing method based on RSS (Responsibility Sensitive Safety). Background Art

[0002] RSS is an interpretable, safety-guaranteed mathematical model that proposes safety verification standards, that is, what are the minimum requirements that each autonomous vehicle must meet and how to verify these requirements. Some scholars have considered the multiple interactions between autonomous vehicles and surrounding vehicles and proposed a new lane change trajectory planning method to enhance the safety constraints of the responsibility-sensitive safety model introduced in the latest 2846-2022-IEEE (Institute of Electrical and Electronics Engineers) standard, thereby improving the safety of the recommended trajectory for a single vehicle. Other scholars have used the extended RSS model to clarify road rights and used a potential field-based sigmoid planner to generate paths, effectively reducing merging time and path length, and eliminating trajectory oscillations, thereby improving the safety and efficiency of autonomous vehicles when merging lanes.

[0003] Reinforcement learning is widely used in the field of autonomous driving decision-making. It provides a flexible and adaptive method that can be used to solve decision-making problems in complex and uncertain environments, and is an important solution in the research of autonomous driving decision-making. Reinforcement learning consists of five elements: agent, environment, state, reward and action. The agent aims to obtain the maximum reward value, interacts with the complex and uncertain environment to learn, perceives the environmental state, action and the resulting reward value, and learns how to take actions to obtain greater rewards through continuous attempts. Actions can not only affect direct rewards, but also affect the next state, and through the next state, affect the subsequent rewards. Some scholars have designed a hierarchical reinforcement learning framework based on deep cyclic deterministic policy gradients, integrated the time-space attention mechanism suitable for image learning driving strategies, improved network combination and efficiency, and completed the lane changing task in the TORCS simulator. Other scholars have proposed an automatic lane changing strategy based on deep reinforcement learning based on proximal policy optimization. This strategy has a greater learning efficiency advantage while maintaining stable performance. Trained autonomous driving vehicles can learn to implement driving policies that are safe, effective and comfortable. Other scholars have comprehensively considered risk assessment and reinforcement learning, and proposed a new risk assessment based on Bayesian theory to quantify driving risks, using deep reinforcement learning to find the strategy with the lowest expected risk, and better complete lane change decisions in static obstacle and dynamic obstacle scenarios in Carla. Other scholars have proposed a safety-based reinforcement learning framework, using the developed safety layer to limit the action space to the subspace of safe actions, solving the problem of lack of safety assurance in reinforcement learning, and using the highD dataset of real highways to complete lane change verification.

[0004] The shortcomings of the autonomous driving decision-making methods in the above-mentioned prior art include: it is often difficult to balance safety and efficiency when facing complex traffic environments, especially when it comes to vehicle interactions, there are certain safety hazards. Existing research on lane changing models ignores the impact of factors such as emergency vehicle priority, driving scene safety, and safety and efficiency balance on autonomous driving lane changing decisions. It rarely considers the reallocation of driving task priorities when facing emergency vehicles and other vehicles with priority rights, and does not systematically evaluate the safety risks of the scene, while ignoring the balance between safety and efficiency. Summary of the invention

[0005] An embodiment of the present invention provides an RSS-based automatic driving right-of-way optimization lane changing method to effectively improve the lane changing safety and efficiency of an automatic driving vehicle in a complex traffic environment.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme.

[0007] An RSS-based automatic driving right-of-way optimization lane changing method includes:

[0008] Establishing a responsibility-sensitive safety model (RSS) model, the RSS model includes a longitudinal minimum safety distance set by using surrounding environment information and vehicle state information of the autonomous driving vehicle AV, and a driving distance safety constraint condition that the longitudinal distance from the surrounding vehicles to the AV is less than the minimum safety distance;

[0009] Constructing a decision-making model, and training the decision-making model using a set global reward function and a vehicle lane-changing performance evaluation index to obtain a trained decision-making model;

[0010] The right-of-way priority strategy is designed by using the trained decision-making model and the driving distance safety constraint condition, and the right-of-way priority strategy is used to make the right-of-way priority lane change decision of the AV according to the surrounding environment information and the right-of-way lane change decision evaluation feedback information.

[0011] Preferably, the RSS model is established, the RSS model includes a longitudinal minimum safety distance set by using the surrounding environment information and vehicle status information of the AV, and a driving distance safety constraint condition that the longitudinal distance from the surrounding vehicles to the AV is less than the minimum safety distance, including:

[0012] The RSS model is established, which includes the longitudinal minimum safety distance d set by using the surrounding environment information and vehicle status information of the AV lon_min , defined as:

[0013]

[0014] Among them, d lon_min Indicates the minimum longitudinal safety distance, v ego represents the longitudinal velocity of AV, α ego represents the longitudinal braking deceleration of AV, v front represents the longitudinal speed of the leading vehicle of the AV, α front represents the longitudinal braking deceleration of the vehicle ahead of the AV, t r Indicates the set reaction time;

[0015] The driving distance safety constraints include longitudinal safety constraints for AV lane keeping, lane changing to the left and lane changing to the right,

[0016] The longitudinal safety constraint for the AV to keep lane is defined as:

[0017] lo sl ≥d lon_min (17)

[0018] Among them, lo ijrepresents the longitudinal distance from the surrounding vehicles to the AV, i=l,s,r represent the left lane, the same lane, and the right lane of the AV, respectively, and j={s,l} represents the vehicle behind and in front of the AV, respectively;

[0019] The AV should meet the longitudinal safety constraints for changing lanes to the left, namely:

[0020]

[0021] The longitudinal safety restraint of the AV changing lanes to the right is defined as:

[0022]

[0023] Preferably, the construction of the decision-making model, using a set global reward function and a vehicle lane-changing performance evaluation index to train the decision-making model to obtain a trained decision-making model, includes:

[0024] Rewards for Interacting AVs with Emergency Vehicles (EMV) emv Designed for:

[0025]

[0026] Among them, emv_follower means that EMV follows AV in the same lane, emv_left_follower and emv_right_follower respectively mean that EMV follows AV in the left or right adjacent lane, emv_far_left and emv_far_right respectively mean that EMV follows AV in the left or right lane, otherwise means that EMV is beyond the perception range of AV.

[0027] The original approximate collision penalty R near Defined as:

[0028]

[0029] The smoothed near collision penalty Defined as:

[0030]

[0031] in, represents the near collision penalty after smoothing, γ represents a predefined constant used to control smoothing, and p t and p t-1 Represent the penalties generated at time t and t-1 respectively;

[0032] The final near collision penalty Defined as:

[0033]

[0034] The safety reward R safe Defined as:

[0035]

[0036] The lane change reward R change Defined as:

[0037]

[0038] Among them, lcc represents the total number of lane changes, and mis_lc represents the number of invalid lane changes;

[0039] Lane Invasion Reward R invasion Defined as:

[0040]

[0041] Among them, la ld and la ego represent the lane boundary and the lateral position of the AV, respectively;

[0042] Will reward R along the center line center Defined as:

[0043]

[0044] Among them, la center Indicates the lateral position of the centerline of the lane where the AV is located;

[0045] The efficiency reward function R eff Defined as:

[0046] R eff =ω time ×R time +ω speed ×R speed (28)

[0047] Among them, R time is the time sub-reward, i.e., the driving time of the AV, R speed is the speed sub-reward, i.e., the difference between the longitudinal speed of the AV and the target longitudinal speed; R eff is the sum of sub-rewards, ω time and ω speed Respectively represent R time and R speed The sub-reward is defined as:

[0048]

[0049] Where t is the AV driving time, v desiredis the target longitudinal speed, i.e. the road speed limit.

[0050] The global reward function is defined as:

[0051] R=R emv +R safe +R change +R invasion +R center +R eff (30)

[0052] The vehicle lane-changing performance evaluation indicators include lane sharing time, lane blocking time and lane blocking percentage. The lane sharing time is the time t required from the EMV starting to follow the AV to the AV successfully giving way to the EMV. s The lane blocking time quantifies the time t during which the AV lane change prevents the EMV from traveling. b , the lane blocking percentage quantifies the percentage of scenarios where the AV changes lanes and hinders the EMV from traveling;

[0053] During the training process of the decision-making model, the values ​​of the global reward function, lane sharing time, lane blocking time, vehicle blocking percentage, average speed and total driving time are obtained. The decision-making model comprehensively considers the values ​​of the global reward function, lane sharing time, lane blocking time, vehicle blocking percentage, average speed and total driving time, adjusts the training parameters, and obtains a trained decision-making model.

[0054] Preferably, the trained decision-making model uses the driving distance safety constraint to design a right-of-way priority strategy, and uses the right-of-way priority strategy to make a right-of-way priority lane change decision of the AV according to surrounding environment information and right-of-way lane change decision evaluation feedback information, including:

[0055] A decision-making model is set up to include a lane-changing information acquisition module and a lane-changing decision algorithm module. The lane-changing information acquisition module determines whether an emergency vehicle exists based on the surrounding environment information of the AV, and inputs the information on whether the emergency vehicle exists and the surrounding environment information of the AV into the lane-changing decision algorithm module. The lane-changing decision algorithm module makes a right-of-way lane-changing decision for the AV based on the information on whether the emergency vehicle exists and the surrounding environment information of the AV, as well as evaluation feedback information using a proximal strategy optimization algorithm. The right-of-way lane-changing decision includes a decision on whether the AV changes lanes and which lane to change to.

[0056] It can be seen from the technical solutions provided by the above-mentioned embodiments of the present invention that the present invention provides a lane changing decision method based on a responsibility-sensitive safety model. The method clarifies responsibility allocation and road right optimization strategies, improves the safety and efficiency of autonomous driving vehicles in complex traffic environments, and can dynamically evaluate and optimize vehicle efficiency, ensure lane changing safety, balance safety and efficiency, and adapt to the road right optimization needs of different vehicles.

[0057] This invention will provide safe, efficient and practical suggestions for autonomous driving vehicle lane changing decisions, and provide a theoretical basis for designing a safety risk assessment framework for autonomous driving scenarios.

[0058] Additional aspects and advantages of the present invention will be given in part in the following description, which will become obvious from the following description, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0060] Figure 1 A model framework diagram of an RSS-based autonomous driving right-of-way optimization lane changing method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0061] The embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0062] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or coupling. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items.

[0063] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.

[0064] To facilitate understanding of the embodiments of the present invention, several specific embodiments will be further explained below with reference to the accompanying drawings, and each embodiment does not constitute a limitation on the embodiments of the present invention.

[0065] The RSS-based automatic driving right-of-way optimization lane changing method proposed in the embodiment of the present invention adopts reinforcement learning to realize the right-of-way optimization strategy of the automatic driving vehicle (Autonomous Vehicle) in the interaction with the emergency vehicle (Emergency Vehicle), realize safe lane changing behavior, and improve the driving efficiency of automatic driving. The model framework of this method is as follows Figure 1 As shown in the figure, the model framework consists of five parts: input data layer, responsibility-sensitive safety module, decision-making module, output layer and feedback module.

[0066] Right of way refers to the priority of autonomous vehicles during driving. Optimizing right of way is critical to the safety and efficiency of autonomous vehicles during driving. Right of way refers to the right of a vehicle to obtain priority in a traffic environment, especially in complex traffic scenarios, such as interactions with emergency vehicles, pedestrians or other vehicles. Optimizing right of way mainly considers the impact of emergency vehicles on autonomous vehicles. Emergency vehicles have the right of way to pass first. When interacting with emergency vehicles, autonomous vehicles must have the ability to identify and respond, give priority to the needs of emergency vehicles, and give way to emergency vehicles in a timely manner.

[0067] Lane changing is one of the important tasks of autonomous vehicles. When changing lanes, autonomous vehicles must take into account multiple factors, including current traffic conditions, the dynamic behavior of other vehicles around them, and their own driving goals and plans to ensure the safety and effectiveness of the decision. Optimizing lane changing decisions can not only improve the driving efficiency of the autonomous vehicle itself and reduce driving time, but also significantly reduce interference with other traffic participants. When lane changing decisions are optimized, vehicles can blend into the traffic flow more smoothly, reducing the impact of sudden braking or acceleration, thereby improving overall traffic fluidity.

[0068] The processing flow chart of a lane changing method for optimizing right of way for autonomous driving based on RSS provided by an embodiment of the present invention is as follows: Figure 1As shown, the process includes the following steps:

[0069] Step S10: Acquire the surrounding environment information and vehicle status information of the AV, where the surrounding environment information includes information such as the speed of surrounding vehicles and the distance from the autonomous driving vehicle;

[0070] Step S20: Establishing an RSS model, which includes the longitudinal minimum safety distance d set by using the above-mentioned surrounding environment information and vehicle status information lon_min , the longitudinal distance from the surrounding vehicles to the AV is less than the above minimum safety distance d lon_min The driving distance safety constraint condition;

[0071] Step S30: Establish a decision-making model, design a road priority strategy using the driving distance safety constraint condition, and implement autonomous learning lane change decision-making in combination with the PPO (Proximal Policy Optimization) algorithm;

[0072] Step S40: Establish an evaluation system and feedback mechanism to evaluate the results of the road right priority decision, and feed back the road right priority decision effect to the decision-making model to improve the decision effect;

[0073] Based on reinforcement learning technology and responsibility-sensitive safety model, a strategy considering the right of way of emergency vehicles is proposed to realize the automatic driving lane changing decision technology.

[0074] The above step S20 includes: the minimum longitudinal safety distance d lon_min Defined as:

[0075]

[0076] Among them, d lon_min Indicates the minimum longitudinal safety distance, v ego represents the longitudinal velocity of AV, α ego represents the longitudinal braking deceleration of AV, v front represents the longitudinal speed of the leading vehicle of the AV, α front represents the longitudinal braking deceleration of the vehicle ahead of the AV, t r Indicates the set response time.

[0077] The above driving distance safety constraints include the longitudinal safety constraints of AV keeping lane, changing lanes to the left and changing lanes to the right,

[0078] The longitudinal safety constraint for the AV to keep lane is defined as:

[0079] lo sl ≥d lon_min (32)

[0080] Among them, loij represents the longitudinal distance from the surrounding vehicles to the AV, i=l,s,r represent the left lane, the same lane and the right lane of the AV respectively, j={s,l} represents the rear vehicle and the front vehicle of the AV respectively, the same below.

[0081] The AV should meet the longitudinal safety constraints for changing lanes to the left, namely:

[0082]

[0083] The longitudinal safety restraint of the AV changing lanes to the right is defined as:

[0084]

[0085] The above-mentioned step S30 includes: setting a decision-making model including a lane change information acquisition module and a lane change decision algorithm module, the lane change information acquisition module determines whether there is an emergency vehicle based on the surrounding environment information of the AV, and inputs the information on whether the emergency vehicle exists and the surrounding environment information of the AV into the lane change decision algorithm module, the lane change decision algorithm module makes a right-of-way lane change decision for the AV based on the information on whether the emergency vehicle exists and the surrounding environment information of the AV, as well as the evaluation feedback information using a proximal strategy optimization algorithm, and the right-of-way lane change decision includes the decision on whether the AV changes lanes and which lane to change to.

[0086] Proximal Policy Optimization (PPO) is a policy optimization method in reinforcement learning, which aims to improve the stability and reliability of training by limiting the policy update amplitude at each step.

[0087] The specific calculation process includes:

[0088] 1. Collect experience data: execute the current strategy to interact with the environment to obtain the state, action, reward, and next state.

[0089] 2. Calculate the advantage function: Estimate the advantage of each state-action pair through GAE.

[0090] 3. Calculate the importance sampling ratio: evaluate the difference between the new and old strategies.

[0091] 4. Apply pruning mechanism: limit the importance sampling ratio to avoid over-update.

[0092] 5. Update strategy and value function: maximize the pruning objective function and optimize the strategy network and value function network.

[0093] 6. Repeat training: optimize multiple times until the strategy converges.

[0094] The above step S40 includes:

[0095] Rewards for AVs interacting with EMVs (Emergency Vehicles) emv Designed for:

[0096]

[0097] Among them, emv_follower means that EMV follows AV in the same lane, emv_left_follower and emv_right_follower respectively mean that EMV follows AV in the left or right adjacent lane, emv_far_left and emv_far_right respectively mean that EMV follows AV one lane away on the left or right side, otherwise means that EMV is beyond the perception range of AV.

[0098] Original approximate collision penalty R near Defined as:

[0099]

[0100] The exponential moving average strategy is introduced to make the near collision penalty smoother in the process of longitudinal distance change, that is:

[0101]

[0102] in, represents the penalty after smoothing, γ represents a predefined constant used to control smoothing, and p t and p t-1 They represent the penalties generated at time t and t-1 respectively.

[0103] Final Near Collision Penalty Defined as:

[0104]

[0105] Safety Reward R safe Defined as:

[0106]

[0107] Lane Change Reward R change Defined as:

[0108]

[0109] Among them, lcc represents the total number of lane changes, mis_lc represents the number of invalid lane changes, and in order to avoid passive lane changes, a constant reward value of 2 is set.

[0110] Lane Invasion Reward R invasion Defined as:

[0111]

[0112] Among them, la ld and la ego denote the lane boundary and the lateral position of the AV, respectively.

[0113] Drive along the center line for R center Defined as:

[0114]

[0115] Among them, la center Indicates the lateral position of the centerline of the lane where the AV is located.

[0116] Efficiency reward function R eff Defined as:

[0117] R eff =ω time ×R time +ω speed ×R speed (43)

[0118] Among them, R time is the time sub-reward, i.e., the driving time of the AV, R speed is the speed sub-reward, i.e., the difference between the longitudinal speed of the AV and the target longitudinal speed; R eff is the sum of sub-rewards, ω time and ω speed Respectively represent R time and R speed The sub-reward is defined as:

[0119]

[0120] Where t is the AV driving time, v desired is the target longitudinal speed, i.e. the road speed limit.

[0121] In summary, the global reward function is defined as:

[0122] R=R emv +R safe +R change +R invasion +R center +R eff (45)

[0123] The present invention uses the following lane-changing performance indicators to comprehensively evaluate the model performance of the method of the present invention.

[0124] 1. Lane sharing time

[0125] This indicator quantifies the time it takes for AV to give way to EMV, that is, the time t required from the time when EMV starts to follow AV to the time when AV successfully gives way to EMV. s .

[0126] 2. Lane blocking time

[0127] This indicator quantifies the time t that AV lane change hinders EMV driving. b .

[0128] 3. Lane Blockage Percentage

[0129] This metric quantifies the percentage of scenarios where the AV changes lanes and hinders EMV driving. This metric is only used for test scenario 2.

[0130] In order to quantitatively evaluate the driving efficiency of AV, the present invention proposes the following indicators.

[0131] 1. Average speed

[0132] This indicator calculates the average speed of AV, the average speed v avg Defined as:

[0133]

[0134] Among them, d lane represents the total driving length of the AV, t d Represents the total travel time.

[0135] 2. Total driving time

[0136] This indicator records the total driving time t of the AV to travel 1000m d .

[0137] The global reward function will provide feedback during the model training process, and the expected training effect can be obtained through the reward function, including road priority and efficiency improvement.

[0138] Lane sharing time and lane blocking time are important indicators for evaluating vehicle right-of-way priority strategies. Vehicle sharing time refers to the total time that the autonomous vehicle is in front of the emergency vehicle and driving in the same lane. Lane blocking time refers to the total time that the autonomous vehicle changes lanes to the same lane as the emergency vehicle and blocks the emergency vehicle. The shorter the lane sharing time and lane blocking time, the better the right-of-way priority strategy of the autonomous vehicle. The length of each training evaluation is fed back to the decision-making model, and the algorithm will follow the feedback information to optimize the right-of-way priority strategy and make better decisions.

[0139] The vehicle blocking percentage indicates the proportion of tests in which the autonomous driving incorrectly changed lanes in front of the emergency vehicle, thus blocking the emergency vehicle from driving. It is used to evaluate the final experimental effect.

[0140] The average speed and total driving time together characterize the driving efficiency of the vehicle. The higher the average speed, the more efficient the vehicle is. The shorter the total driving time, the more efficient the vehicle is. The decision-making model will adjust its speed strategy based on these two indicators to improve driving efficiency.

[0141] During the training process of the decision-making model, the values ​​of the global reward function, lane sharing time, lane blocking time, vehicle blocking percentage, average speed and total driving time are obtained. The decision-making model comprehensively considers the values ​​of the global reward function, lane sharing time, lane blocking time, vehicle blocking percentage, average speed and total driving time, adjusts the training parameters, and obtains a trained decision-making model.

[0142] In summary, the present invention proposes an automatic driving right-of-way optimization lane changing method based on the RSS model, which can effectively improve the lane changing safety and efficiency of automatic driving vehicles in complex traffic environments. It aims to solve the above challenges of existing automatic driving technology in complex traffic environments, dynamically evaluate and optimize vehicle efficiency, ensure lane changing safety, balance safety and efficiency, and adapt to the right-of-way optimization needs of different vehicles.

[0143] The RoW-RL model proposed in the present invention comprehensively considers the driving safety of autonomous vehicles, the right of way priority of emergency vehicles and the driving efficiency of autonomous vehicles, and realizes a lane-changing method with optimized right of way. The RoW-RL model proposed in the present invention is based on the responsibility-sensitive safety model and combines the design of the minimum safe longitudinal distance to propose safety constraints for the lane-changing process.

[0144] This invention will provide safe, efficient and practical suggestions for autonomous driving vehicle lane changing decisions, and provide a theoretical basis for designing a safety risk assessment framework for autonomous driving scenarios.

[0145] Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0146] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention or certain parts of the embodiments.

[0147] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0148] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A lane changing method for optimizing right of way for autonomous driving based on RSS, characterized in that: include: Establishing a responsibility-sensitive safety model (RSS) model, the RSS model includes a longitudinal minimum safety distance set by using surrounding environment information and vehicle state information of the autonomous driving vehicle AV, and a driving distance safety constraint condition that the longitudinal distance from the surrounding vehicles to the AV is less than the minimum safety distance; Constructing a decision-making model, and training the decision-making model using a set global reward function and a vehicle lane-changing performance evaluation index to obtain a trained decision-making model; The right-of-way priority strategy is designed by using the trained decision-making model and the driving distance safety constraint condition, and the right-of-way priority strategy is used to make the right-of-way priority lane change decision of the AV according to the surrounding environment information and the right-of-way lane change decision evaluation feedback information.

2. The method according to claim 1, characterized in that The RSS model is established, the RSS model includes a longitudinal minimum safety distance set by using the surrounding environment information and vehicle status information of the AV, and a driving distance safety constraint condition that the longitudinal distance from the surrounding vehicles to the AV is less than the minimum safety distance, including: The RSS model is established, which includes the longitudinal minimum safety distance d set by using the surrounding environment information and vehicle status information of the AV lon_min , defined as: Among them, d lon_min Indicates the minimum longitudinal safety distance, v ego represents the longitudinal velocity of AV, α ego represents the longitudinal braking deceleration of AV, v front represents the longitudinal speed of the leading vehicle of the AV, α front represents the longitudinal braking deceleration of the leading vehicle of the AV, t r Indicates the set reaction time; The driving distance safety constraints include longitudinal safety constraints for AV lane keeping, lane changing to the left and lane changing to the right, The longitudinal safety constraint for the AV to keep lane is defined as: it sl ≥d lon_min (2) Among them, lo ij represents the longitudinal distance from the surrounding vehicles to the AV, i=l,s,r represent the left lane, the same lane, and the right lane of the AV, respectively, and j={s,l} represents the vehicle behind and in front of the AV, respectively; The AV should meet the longitudinal safety constraints for changing lanes to the left, namely: The longitudinal safety restraint of the AV changing lanes to the right is defined as:

3. The method according to claim 2, characterized in that The construction of the decision-making model, training the decision-making model using the set global reward function and the vehicle lane-changing performance evaluation index to obtain a trained decision-making model, includes: Rewards for Interacting AVs with Emergency Vehicles (EMV) emv Designed for: Among them, emv_follower means that EMV follows AV in the same lane, emv_left_follower and emv_right_follower respectively mean that EMV follows AV in the left or right adjacent lane, emv_far_left and emv_far_right respectively mean that EMV follows AV in the left or right lane, otherwise means that EMV is beyond the perception range of AV. The original approximate collision penalty R near Defined as: The smoothed near collision penalty Defined as: in, represents the near collision penalty after smoothing, γ represents a predefined constant used to control smoothing, and p t and p t-1 Represent the penalties generated at time t and t-1 respectively; The final near collision penalty Defined as: The safety reward R safe Defined as: The lane change reward R change Defined as: Among them, lcc represents the total number of lane changes, and mis_lc represents the number of invalid lane changes; Lane Invasion Reward R invasion Defined as: Among them, la ld and la ego represent the lane boundary and the lateral position of the AV, respectively; Will reward R along the center line center Defined as: Among them, la center Indicates the lateral position of the centerline of the lane where the AV is located; The efficiency reward function R eff Defined as: R eff =ω time ×R time +oh speed ×R speed (13) Among them, R time is the time sub-reward, i.e., the driving time of the AV, R speed is the speed sub-reward, i.e., the difference between the longitudinal speed of the AV and the target longitudinal speed; R eff is the sum of sub-rewards, ω time and ω speed Respectively represent R time and R speed The sub-reward is defined as: Where t is the AV driving time, v desired is the target longitudinal speed, i.e. the road speed limit. The global reward function is defined as: R=R emv +R safe +R change +R invasion +R center +R eff (15) The vehicle lane-changing performance evaluation indicators include lane sharing time, lane blocking time and lane blocking percentage. The lane sharing time is the time t required from the EMV starting to follow the AV to the AV successfully giving way to the EMV. s The lane blocking time quantifies the time t during which the AV lane change prevents the EMV from traveling. b , the lane blocking percentage quantifies the percentage of scenarios where the AV changes lanes and hinders the EMV from traveling; During the training process of the decision-making model, the values ​​of the global reward function, lane sharing time, lane blocking time, vehicle blocking percentage, average speed and total driving time are obtained. The decision-making model comprehensively considers the values ​​of the global reward function, lane sharing time, lane blocking time, vehicle blocking percentage, average speed and total driving time, adjusts the training parameters, and obtains a trained decision-making model.

4. The method according to claim 1, characterized in that The trained decision-making model uses the driving distance safety constraint to design a right-of-way priority strategy, and uses the right-of-way priority strategy to make a right-of-way priority lane change decision for the AV based on surrounding environment information and right-of-way lane change decision evaluation feedback information, including: A decision-making model is set up to include a lane-changing information acquisition module and a lane-changing decision algorithm module. The lane-changing information acquisition module determines whether an emergency vehicle exists based on the surrounding environment information of the AV, and inputs the information on whether the emergency vehicle exists and the surrounding environment information of the AV into the lane-changing decision algorithm module. The lane-changing decision algorithm module makes a right-of-way lane-changing decision for the AV based on the information on whether the emergency vehicle exists and the surrounding environment information of the AV, as well as evaluation feedback information using a proximal strategy optimization algorithm. The right-of-way lane-changing decision includes a decision on whether the AV changes lanes and which lane to change to.

Images