The Obstacle is the Path: Designing a Pop-Up Obstacle for Vehicle Collision System Testing

Transcription

1 The Obstacle is the Path: Designing a Pop-Up Obstacle for Vehicle Collision System Testing Designers: Mentors: Advisor: Zach Chase Richard Lui Avinash Balachandran Joe Funke Chris Gerdes

2 Page 1 Opener Our research process is based on the scientific method that Stanford s Institute of Design calls Design Thinking. This method allows us to distill core needs that our safety technology should address and iterate quickly towards a functional end device. We used this process to not only design our safety technology but also to present our findings in this paper. We hope you will enjoy delving into our report. FIGURE 1: The Design Thinking Method from dschool.stanford.edu

3 Page 2 Empathize/Motivation According to NHTSA s Traffic Safety Fact Sheet, 5,615,000 police-reported traffic crashes occurred in Better perception and vehicle control technologies can reduce crash statistics with systems that help avoid collisions. The Insurance Institute for Highway Safety has seen a 7 percent reduction in crashes for vehicles with a basic forward-collision warning system, and a 14 to 15 percent reduction for those with automatic braking. 2 Current research explores additional safety improvements with further automation. Shared control strategies enable the driver to operate the vehicle within safe envelopes, but deviates from driver commands to avoid collisions if necessary. 3 Fully autonomous vehicles can operate without human intervention and can be designed to react in emergency situations, such as conducting an emergency lane change to get out of an unsafe lane. 4 However, shared and fully autonomous obstacle avoidance systems must be rigorously benchmarked before being tested on open roads. Existing obstacle platforms for autonomous and semi-autonomous vehicle testing exist, 5 but are either prohibitively expensive or too limited in their capabilities. Obstacle avoidance testing currently lacks cost-effective yet dynamic obstacles that can pop-out, for example, to model emergency situations. Our team has developed a low-cost obstacle for validating highly automated and autonomous obstacle avoidance systems. This experimental device simulates imminent collision situations like a car suddenly pulling out of a hidden driveway or a deer jumping into the road. Our test setup will assist in validation studies of obstacle avoidance systems, which will ultimately improve the safety of the entire passenger fleet and could eventually mitigate traffic accidents Balachandran, Avinash, Stephen M. Erlien, and J. Christian Gerdes. "The Virtual Wheel Concept for Supportive Steering Feedback During Active Steering Interventions." ASME 2014 Dynamic Systems and Control Conference. American Society of Mechanical Engineers, Funke, Joseph, and J. Christian Gerdes. "Simple clothoid paths for autonomous vehicle lane changes at the limits of handling." ASME 2013 Dynamic Systems and Control Conference. American Society of Mechanical Engineers,

4 Page 3 Define We hope to see collision avoidance systems in production cars around the world. Such systems themselves need to be rigorously tested and validated using obstacles to avoid. We identified several key requirements, outlined below, necessary for such an obstacle design to be useful for testing collision avoidance systems. Design Requirement Activates in less than 1 second Easy triggering Collision safe Obstacle blocks a lane Low cost (<$150) Simple setup (<1 hour and few steps) Low reset time (<30 seconds) Portability to test site Reason The obstacle must deploy from an inactive to an active state in order to test last-minute obstacle avoidance systems. This simulates real-world scenarios such as a jumping deer or previously unseen vehicle. Quick activation tests response times of avoidance systems and drivers. The obstacle should offer easy methods of triggering that can be both repeatable and easy to change during testing. During the testing and validation phase, avoidance systems will inevitably fail to avoid the obstacle in some situations. Therefore, the obstacle must be both robust to collisions with the vehicle and not damage the vehicle. This must block a lane so that it can initiate a driver or system to perform a lane change. Obstacles must be inexpensive obstacles so that they can be replicated, enabling multi-obstacle tests, and easily replaced if damaged or otherwise worn out. A simple setup reduces test preparation and supports potential volunteer assistance with complex experimental setups. Low time to reset the obstacle between deployments enables quick repetition of tests, which is especially critical for time-critical shared control experiments with volunteer drivers. Multiple obstacles must fit in one large car.

5 Page 4 Ideate During ideation, we developed various initial designs that could potentially meet the design requirements. Families of ideas involved hydraulic pressure, springs and levers, compressed air, water, and light. Figure 2 illustrates 3 preliminary design sketches based on these ideas. a) b) c) Figure 2: Initial design sketches. In a), a mousetrap-inspired obstacle uses two spring-loaded poles to quickly lift a flag obstacle. In b), small air blowers create a wall of tubes. In c), a single larger blower fills a tube with a catch mechanism at the end to prevent it from dancing in the wind. From the different families of ideas, an air-based inflation obstacle fulfilled the design requirements best while remaining inexpensive to build. An off-the-shelf boat blower (see Figure 3) offered an inexpensive way to inflate an obstacle. The result was a cheap, easy to control in an experimental setup, quick to reset between experimental trials, and crash-safe.

6 FIGURE 3: A boat blower, an inexpensive fan used to move air around a boat s cabin. Page 5

7 Page 6 Prototyping Prototypes enabled quick iteration through potential designs, allowing us to identify and improve on challenges with each new iteration. We prototyped both the form of the obstacle and the triggering mechanism for obstacle deployment. Obstacle Form Initial prototypes incorporated an electric boat blower duct-taped to a plastic garbage bag (see Figure 4). This setup confirmed that potential deployment speeds and obstacle lengths were within reach of our requirements and allowed optimization of tube length, tube pre-inflation shape, tube post-inflation shape, fan ramp-up time, and thrust vectoring to minimize activation time. We ultimately determined we needed greater air velocity and momentum than the blower could provide; at best it could only inflate a 4 foot tube with in activation time slightly over one second. Figure 4: A prototype test with a paper cone nozzle to increase air velocity within the tube. Additional prototypes compared air compressors and other, higher volume blowers and ultimately settled on a small and cheap handheld electric leaf blower (see Figure 5). With further tube design iteration this blower inflated a 10 foot long tube in under half a second, making it far superior to the boat blower setup. Tube design was primarily focused on reducing inflation time, minimizing wear and tear, and enabling manufacturability, which led to a simple open-ended tube design. Our final tube design was sewn together from red table cloth, which we found to be both lightweight and durable. Other components such as visibility and stability were also considered. We tested increased visibility with caution streamers, for example, but found even this incremental weight increase pulled the tube down too much. We also noticed issues with instability, in which the tube would flap around even in low wind. Cutting three slits in the end of the tube to mimic the effect of a diffuser greatly improved stability.

8 Page 7 Figure 5: We tested the leaf blower with different inflatable shapes to determine the cheapest solution. The streamers added visibility for the driver but weighed down the tube too much. The blower is mounted on an off-the-shelf sawhorse, which is inexpensive, durable, and stackable. Even with the necessary electronics built in, each sawhorse can be stacked, which was critical for transporting multiple obstacles to a testing facility (see Figure 6). FIGURE 6: Three obstacles, stacked and ready for transportation.

9 Page 8 Obstacle Trigger The obstacle needs to be triggered in some way. Ideally, triggering should be correlated with the test vehicle s spatial position. For example, a common triggering scenario would be to have the obstacle deploy when the vehicle is within some distance of the obstacle. There are many ways this can be accomplished. Hydraulic pressure hoses, which are triggered when a vehicle drives over them, are commonly used in city streets to count cars. We considered these systems as well as laser trip wires, sonar sensors, and GPS-based wireless triggers. The pressure hose was expensive and the laser trip wire implementation was susceptible to environmental conditions. Therefore, we built the following two obstacle triggering mechanisms: 1. Sonar sensor triggering mechanism 2. GPS-based wireless triggering mechanism Both of these mechanisms use the popular and inexpensive Arduino platform as the computation backbone. Sonar sensor triggering mechanism Figure 7 illustrates the sonar tripwire concept. When the vehicle drives past this tripline, the obstacle triggers. The sonar sensor was built into a small tower placed just outside the lane (see Figure 8). When a vehicle drives by the tower, it triggers the obstacle, causing the tube to inflate. Initial prototypes helped determine the range and reliability of detecting vehicles, and small variations were tested to improve performance. FIGURE 7: Sonar sensor data monitored by an Arduino controls whether the blower is on.

10 Page 9 Figure 8: (left) First prototype of the sonar sensor tower. (right) Final iteration of the sonar sensor tower. The sonar sensor offered straightforward integration with the Arduino platform and low cost. The reliable sensor range was found to be 8 feet. This mechanism is also independent of GPS coverage, which is a crucial advantage over the GPS-based triggering system. However, sensor tower placement places limitations on the kinds of experimental setups that can be created with this mechanism. Furthermore, because the sonar towers need to be physically wired to the obstacle themselves, wiring complicated experimental setups. These issues were overcome by also creating a GPS-based wireless triggering system. GPS-based wireless triggering mechanism Figure 9 illustrates the GPS-based wireless triggering mechanism. The concept behind this mechanism is that a computer on-board the vehicle determines when to trigger the obstacle based on the GPS position of the vehicle. This triggering command is then wirelessly sent to the obstacle, causing it to deploy. FIGURE 9: In this setup, a computer onboard the vehicle monitors vehicle position from the vehicle s GPS, and sends a trigger command through a pair of wireless Arduinos to trigger the blower.

11 Page 10 The main benefit of this mechanism is the triggering range. The wireless range of the RF shields were found to be about 75 [m]. For vehicles travelling at highway speeds, the obstacle could trigger at least 2.5 [s] before the vehicle reached the obstacle itself. This was crucial for testing collision avoidance systems meant for highway use. This mechanism was easier to set up than the sonar mechanism as well, since the towers were unneeded. The main drawback of this system is that it depends on good GPS coverage. The two triggering systems complement each other well, enabling a variety of different experimental setups. Figure 10 illustrates the complete pop-up obstacle solution. FIGURE 10: Back view of the final obstacle design, with components called-out.

12 Page 11 Testing Tests conducted in conjunction with prototyping confirm fulfillment of the initial design requirements. Design Requirement Result Activates in less than 1 second Video playback confirm an average inflation time of 0.3 seconds. Easy triggering Collision safe Obstacle blocks a whole lane Low cost (<$150) Simple setup (<1 hour and few steps) Low reset time (<30 seconds) Portability to test site Sonar sensor towers and GPS-based wireless triggering provide two complementary triggering options. No damage to either the tube or vehicle occur on collision. An obstacle tube length of 10 feet blocks a lane. A single obstacle costs $132 to build. Setup takes about 30 minutes and involves a 3 step assembly process. A reset takes about 30 seconds for a first time user and decreases with experience. Stackable components enables transportation of multiple obstacles. The obstacles have also been used in tests of collision avoidance systems. These tests were conducted using X1 (see Figure 11). FIGURE 11: X1, a student-built, all-electric, drive-by-wire research testbed.

13 Page 12 Shared Control Collision Avoidance We used the obstacles to create an obstacle course for shared control collision avoidance systems. Research participants were tested on their ability to avoid pop-up obstacles with and without haptic feedback from the car. Figure 12 illustrates the experimental setup used in this test. Figure 12(a) shows the driver in a two lane road with 3 pop-up obstacles (dotted red dots) before an obstacle deployment. Figure 12 (b) illustrates one of these obstacles randomly deploying, causing the driver to perform a lane change maneuver. Figure 13 illustrates this from the driver perspective. As expected, drivers performed better with the feedback while sharing control with the obstacle avoidance system. FIGURE 12: Shared control experiment. Three obstacles were present, and one could trigger. The driver attempted to avoid the obstacle, possibly with the assistance of the shared controller. FIGURE 13: (left) Driver does not know which of three obstacles will deploy. (right) The third obstacle deploys, and the driver attempts to avoid it. Autonomous Control Obstacle Avoidance We also used the obstacles to model emergency situations for autonomous obstacle avoidance. Here the autonomous vehicle followed a predefined path, and an obstacle deployed, necessitating evasive maneuvers from the controller. These tests enabled evaluation of how well autonomous controllers react to obstacles that appear on the road. Figure 14 depicts a lane change setup, and Figure 15 demonstrates an obstacle popping up mid-turn.

14 Page 13 FIGURE 14: (left) The autonomous vehicle drives in its lane and (right) swerves to avoid the obstacle that pops out. FIGURE 15: (left) The autonomous vehicle drives around a turn. (right) The obstacle deploys and the vehicle swings wide to avoid it.

15 Page 14 Conclusion The testing and validation of obstacle avoidance systems for autonomous and shared control vehicles are pressing safety issues. Some testing and validation can be done through simulation; however, there is no substitute for experiments using real vehicles in real-world conditions. In these scenarios, having an experiment that can simulate an obstacle suddenly appearing, or popping up, is very helpful. Such an obstacle must be easy to setup and trigger, quick to trigger, portable and inexpensive. Also, since there will be scenarios where a collision avoidance system will invariably fail, this obstacle must withstand a collision without causing major damage to the vehicle or obstacle setup. Our product meets these needs. In addition to meeting these needs, our design has been employed in actual experiments meant to test collision avoidance systems. Results from the shared control experiments demonstrate improved driver reaction times with the collision system in place, which resulted in less aggressive, more comfortable maneuvers. Results from the fully autonomous tests show that even as an autonomous vehicle approached its handling limits, its collision avoidance system was still capable of avoiding the pop-up obstacle. Both of these experimental setups, enabled by the pop-up obstacle system, are leading to improved obstacle avoidance systems for future vehicles. Tests such as these, made possible with the pop-up obstacle, will lead to improved collision avoidance technologies for the entire passenger vehicle fleet. One day, these kinds of technologies could lead to accident-free transportation.