Invented by Leo Nelson Chan, Kristopher Keith Gaudin, Roxane Lyons, William J. Leise, John A. Nepomuceno, Rajiv C. Shah, Edward P. Matesevac, III, Jennifer Criswell Kellett, Steven Cielocha, Jeremy Myers, Matthew S. Megyese, Jennifer L. Crawford, State Farm Mutual Automobile Insurance Co
The State Farm Mutual Automobile Insurance Co invention works as followsSystems and Methods are disclosed for adapting autonomous vehicle driving behaviour. An aggression factor may be used by a controller to control the autonomous vehicle. The autonomous vehicle can detect and analyze one or more characteristics from nearby vehicles to estimate the aggression level. It then adjusts the aggression factor according to the estimated level.
Background for System for adjusting autonomous vehicle driving behavior to mimic that of neighboring/surrounding vehicles
When driving, both the driver and passenger assume some risk of personal injury or damage to property. The risk can be reduced or eliminated by eliminating certain contributing factors. A driver can avoid risky behaviors such as driving drunk, driving tired or texting while driving. Another example is that a driver can reduce the risk of injury by driving an automobile with safety features like airbags and seatbelts.
However certain risk factors might not be reduced. The very nature of the vehicle can present inherent risks. A car can weigh up to thousands of pounds, and it may be difficult to maneuver or stop. Even at moderate speeds, a collision can cause serious damage to a vehicle and injury to its occupants. A driver can also be at risk from the biggest factor in driving, other drivers.
In some cases, environmental factors can influence the relative safety or riskiness of a place. A driver may not notice a bridge that is only one lane in a valley between hills until they have crested the hill. The driver will have less time to react when a second vehicle approaches the bridge in the opposite direction if the distance between the hill and bridge is small. The driver has little or no control over environmental factors.
Moreover, environmental elements that contribute to the riskiness in an area are not always readily visible, observable or quantifiable. Even if a civil engineering identifies several one-lane highway bridges as potentially hazardous, she may not be able to quantify how dangerous these bridges are in relation to each other. The engineer might also overlook a seemingly safe two-lane road that in reality is more dangerous than the one-lane roads. These environmental risk factors can go unnoticed because the factors that contribute to risk are not always obvious, observable or quantifiable. Engineers and government officials might never be able to identify high-risk areas.
This specification describes systems and methods to identify the riskiness of certain areas (such a roads, intersections and bridges, road segments and parking lots) for vehicles and their occupants. A risk index can be calculated, in particular, for a variety of areas. The risk indexes can be compared, which allows a comparison to the relative riskiness. Prioritizing the riskiest areas is a good way to categorize them. Prioritizing construction projects is especially important when funds are limited. The disclosed systems and method can be used to identify the roads, intersections or bridges that need redesigning or additional construction.
In one aspect, an computer-implemented technique may include calculating the risk index. Calculating the index can include one or more of the following: (i), calculating an expected number of collisions over a period of time; (ii), determining the number observed collisions in a specific area during the period of time; (iii), estimating the aggression level for a vehicle nearby; and/or, (iv), calculating the index based on a comparison of the expected number of collisions with the observed number of collisions; and/or the estimated aggression of the vehicle nearest. The number of anticipated collisions can be calculated using (a) historic traffic data in the area or (b) traffic data from multiple areas, which corresponds to the average risk across the multiple areas. In another embodiment, the expected number of collisions can be calculated based on traffic flow or volume, and/or adjusted to account for market penetration. The observed collisions can be restricted to only observations of vehicles in the market that corresponds to the market penetration. In one embodiment, generating a map of risk based on the risk index is also included in the method. The risk map can include graphic elements that depict risk indices in multiple areas, as well as the specific area. The risk map can be a heatmap, or it may be sent to mobile devices by wireless communication or via data transmission through one or more radio channels or wireless communication channels.
The instructions may cause the processor do any or all of the following: (i) calculate a number of expected collisions over a time period; (ii) determine a number observed collisions for a particular area over a time period; (iii) estimate an aggression level of a proximate vehicle; and/or,(iv), calculate a risk indice based upon a comparison between the number anticipated collisions and a number observed. The instructions can cause the processor do one or more of these things: (i), calculate a total number of collisions expected over a period of time; (ii), determine the number observed collisions in a specific area during the period of time; (iii), estimate the aggression level for a vehicle nearby; and/or, (iv), calculate a risk indice based on a comparison of the total number of collisions expected over a period of time and the number observed collisions. In one embodiment, instructions can cause the processor calculate the number expected collisions using historical traffic data. The instructions can cause the processor calculate the number expected collisions using historical traffic data from multiple areas. This number may be a risk average for those multiple areas. The instructions can cause the processors to calculate the expected number of collisions based on traffic flow and/or adjust that number for market penetration. The instructions can cause the processor only to observe collisions that involve vehicles in the market according to the market penetration. Or, the instructions can cause the processor generate a map of risk based on the risk index. In one embodiment, instructions can cause the processor generate a map of risk indexes in multiple areas, including the specific area. The map may also be generated as a heatmap. The system can include a display for displaying the map and/or transmitting the map to another device or system for display. This could be done via wireless communication, data transmission through one or more radio channels or wireless communication channels, or even a heat map.
In another aspect, the computer system can include a method for calculating risk index. In one embodiment, a computer system may include a means to calculate a risk index. The means can be any of: (i), means to calculate a number expected collisions during a given time period (ii), means to determine a number observed collisions in a specific area for the same time period (iii), means for estimation of an aggression level for the vehicle adjacent; or (iv), means for computing the risk index using a comparison of the expected collisions with the observed collisions. In one embodiment, a system can include means to calculate the number of anticipated collisions using historical traffic data. In another embodiment of the system, a method for calculating expected collisions using historical traffic data from multiple areas may be included. The number of anticipated collisions can be a measure of the average risk in multiple areas. The system can include a way to calculate the number expected collisions based on traffic flow and/or a way to adjust the number expected collisions based on market penetration. The system can include a way to limit the observed collisions only to those involving vehicles in the market that corresponds to the market penetration. It may also include a way to generate a map of risk based on the risk index. The system can also include graphic elements that show risk indices in multiple areas, as well as the specific area. The system can include a way to generate the heat map as a risk map.
The system may also include a sensor mounted on an autonomous car that detects the characteristics of a nearby vehicle, and a controller that controls the autonomous car to (i) estimate the aggression level of that vehicle by analyzing the characteristic; (ii), adjust an aggression coefficient based upon the estimated aggression; and (iii), control the autonomous vehicles using the aggression factor. Sensors include motion sensors, audio sensors, and laser sensors. The characteristic can be acceleration, speed and/or a factor of swerve. The controller can be set up to change the rate of acceleration and deceleration based on an aggression factor.
In another aspect, the method of controlling an autonomous car may include: detecting a characteristic from a nearby vehicle using a sensor attached to the vehicle; analysing the characteristic to determine an aggression level for the vehicle; adjusting a factor of aggression based on this estimated level; and controlling autonomous vehicle by utilizing the factor of aggression. The characteristic can be detected by detecting the motion of the vehicle or audio associated with it. The characteristic can be speed, acceleration and/or a steering factor. The controller may adjust the rate of acceleration and deceleration based on an aggression factor.
The system may also include a method for adjusting the aggression factor according to the estimated level of aggression. The sensor may be a motion detector, an audio or laser sensor, or a sensor for audio. “The means of controlling the autonomous car using the aggression factor can include adjusting a rate at which a controller controls the autonomous vehicle, based on this aggression factor.
This summary introduces a number of concepts which are described in detail below. Feature and advantages that are not included in this summary may be present in certain embodiments. Further, some embodiments could omit all or part of the advantages and features described in this summary.
This specification describes systems and methods to calculate a risk index in one or more areas, such as roads, intersections and bridges, parking lot segments, portions of road, or other transportation infrastructure. Comparing risk indices of different areas allows for a comparison of their relative risks. These risk indices can be used to assess the relative riskiness of different transportation infrastructures, and/or identify infrastructures that need safety improvements. In some embodiments, for example, a map of risk indexes may be created to visually represent one or more areas in a region. The risk map can be used to identify the most risky areas in the region.
Exemplary method 100 for Calculating Risk Index
FIG. FIG. 1 shows an example computer-implemented risk calculation method 100 according to a particular embodiment. The system 200 in FIG. can implement the method 100 in its entirety or part. 2. “The method 100 can be stored in a memory, as one or several instructions or routines.
The method 100 can begin after collisions are predicted (block 102). Calculating expected collisions can be done using historical traffic data. In one embodiment historical traffic data for an area of interest (including historic auto-insurance claim data and/or collision data for vehicles) may be analyzed. An average number of accidents may then be calculated for the expected collisions. As an example, historically area x has averaged 10 collisions per monthly. The expected collisions in area x could be 10 collisions a month. In some embodiments historical traffic data from multiple areas can be analyzed and the average number collisions for each area calculated and used to calculate the expected collisions. As an example, historically five different areas, including area x, may have averaged a total 62 collisions each month. The expected collisions in any area (e.g. area x, for example) within the region may be 12.4 collisions per monthly.
In some embodiments, expected collisions can be a function traffic volume. A collision rate per 100 traffic vehicles could be calculated, for example, using historical data (such as auto insurance claims data or vehicle collision data), for a specific area (e.g. 5.5 collisions for every 100 traffic vehicles) or multiple areas. The traffic volume in the area of concern (e.g. area x), can be observed (e.g. 200 vehicles per monthly). Next, you can calculate the collisions expected for the area (e.g. area x may expect 11 collisions per month). In one embodiment, market penetration may be used to adjust the number of collisions expected. If an insurance company has a 25% market share, it may calculate the number of expected collisions as 2.75 (i.e. 25% of 11), which represents the expected number collisions involving vehicles insured by that insurance company. Another example is that a vehicle manufacturer can adjust for market share to determine the expected number of accidents involving vehicles of a certain make and/or type.
Real collisions can be observed” (block 104) Data from insurance claims can be used to identify collisions at the area in question. An analysis of the insurance claims data could reveal, for example, that 15 insurance claims were filed over a period of one month on collisions that occurred in area x. Comparing the area x collision rate to the 11 collisions calculated in the previous example, the collision rate was higher than expected.
A risk-index can be calculated by comparing expected collisions with observed collisions” (block 106). The number of observed collisions can be divided by the expected collisions. A risk index greater than 1 indicates that an area is more risky than expected. The numbers in the above example can be used to determine that area x has a risk of 1.36. (i.e. the result of 15/11).
A risk map can be generated, at least partially, based on the calculated index of risk (block 108). The system 200 may display the risk map. The risk map can include graphical elements that represent risk indices in areas included on the map. The risk map, for example, may include colored rings indicating a calculated risk index associated with areas included on the map.
The calculated risk indices can be compared with one another to enable an evaluation of relative riskiness in the areas associated with these risk indices. This evaluation can be performed easily by the user using the generated risk map. This risk assessment may be particularly useful to civil engineers and government officials who are interested in identifying infrastructure that needs safety improvements. The method 100 can also be helpful to insurance companies. Insurance companies, for example, may adjust their rates depending on how frequently a driver passes through an area with a higher risk index. The rate adjustment can be based on an estimate or implemented as part a dynamic policy. An insurance company, for example, may use a dynamic rate that responds to the real-time behavior of a driver and his routing to reward drivers who are risk averse. The rate could go up or lower dynamically as the driver passes through zones with high or low risk indexes.
Click here to view the patent on Google Patents.