Daedalus Autonomous Vehicle

Transcription

1 Daedalus Autonomous Vehicle June 20, 2002 Team Members: Nicole Anthony Byron Collins Michael Fleming Chuck Liebal Michelle Nicholas Matthew Schmid Required Statement from Faculty Advisor I, Dr. Charles Reinholtz of the Department of Mechanical Engineering at Virginia Polytechnic Institute and State University, do hereby certify that the engineering design of the new vehicle, Daedalus, has been significant and each senior team member has earned six semester hour credits for their work on this project. Signed, (Date) Dr. Charles Reinholtz (540)

2 Introduction The Virginia Tech Autonomous Vehicle Team (AVT) and Team Daedalus is proud to present a new vehicle for entry in the 10 th Annual Intelligent Ground Vehicle Competition (IGVC). The Daedalus Autonomous Vehicle will compete in all three challenges of the competition: Vehicle Design, Autonomous Challenge, and the Navigation Challenge. The team implemented a structured, systematic design process, and the latest technologies in the design of this vehicle. Daedalus is named after a mythological Greek inventor who created a labyrinth and was the only one who could escape the maze. Design Process The design process, as described in Product Design and Development (Ulrich 2000), is shown in Figure 1. The steps in this process are iterative, meaning that one or more steps may be repeated until satisfactory results are obtained. Mission Statement Identify Customer Needs Establish Target Specifications Generate Product Concepts Test Product Concept(s) Set Final Specifications Plan Downstream Development Development Plan Figure 1. Schematic of the design process. The mission statement provides a working foundation by outlining the scope of the project. The mission statement includes a product description, key business and educational goals, market definition, assumptions, and stakeholders. The product description characterizes the goal without specifying a detailed design concept. Team Daedalus s product description is a vehicle that is able to navigate autonomously and adheres to the rules of the 10 th Annual Intelligent Ground Vehicle Competition. Some of our key goals include designing a competitive vehicle and effectively using the design process in a team environment. The team targets two groups of customers: governmental and educational. Our governmental customers include the United States Military and the Department of Transportation, and our educational customers are our faculty advisor, Dr. Charles F. Reinholtz and ourselves. Our primary stakeholders are the team sponsors, Virginia Tech, and the design team. 2

3 After defining the mission statement, the team identified a set of customer needs. The first step in this phase was to interview the primary and secondary customers as defined in the mission statement. Dr. Reinholtz, former AVT members, IGVC judges, and team members from other universities were consulted. During these interviews, questions were asked concerning previous vehicle performance, design flaws, and events at previous competitions. The results of these interviews, as well as information from competition rules, set a foundation for the generation of customer needs. Each customer need was numerically ranked from least important to most important. This ranking system focused the team on the most important customer needs. After identifying customer needs, target specifications were set. The first step in this phase was developing a list of metrics. Metrics are measurable characteristics of a product, such as weight or maximum speed, which quantify the customer needs. Once the list of metrics was developed, previous vehicles were benchmarked using the metrics as guidelines for comparison. On-line design reports from prior competitions provided an excellent reference for gathering competitive benchmarking information. The following vehicles were studied for their relative strengths and weaknesses: Virginia Tech Maximus (2001), Virginia Tech Navigator (2000), Virginia Tech Artemis (1999), West Point MAGIC (2001), and Hosei University Amigo (2001). By comparing the characteristics of these other vehicles, the team was able to set target specifications for Daedalus. Conceptual designs were then generated based on target specifications. Figure 2 shows an AutoCAD Mechanical Desktop 6.0 rendering of one of the early concepts. The team then reflected on the design process and results to ensure practicality and team agreement. By taking a systematic approach throughout vehicle development, the team was able to target specific problems inherent in previous designs and develop solutions for them. VT Maximus had an unstable frame and unorganized components. VT Navigator was an oversized vehicle that incorporated older technologies and components. VT Artemis was difficult to maintain. All of these vehicles had detached hoods which were cumbersome and heavy. The conceptual design for Daedalus focused first on creating a stable, easily accessible, well-organized, and compact vehicle. Figure 2. CAD Rendering of Daedalus Conceptual Design Overall cost, feasibility, and adherence to target specifications were some of the criteria used for concept selection. Concepts were scored and ranked. The highest-ranking concept, a compact three-wheeled design, was chosen for further 3

4 development. With the concept clearly defined, the team established more detailed cost estimates and began to procure resources to build and test a prototype. A bill of materials was created, which summarized both fixed and variable costs. Eliminating complexity and unnecessary parts, using different materials, and optimizing parts reduced the cost and complexity of the final design. The team actively sought donations for expensive components as well as solicited price quotes from multiple suppliers. Using standard components such as aluminum channel, tubing, and angle, rather than specially machined components, the time and cost required to create these custom components was eliminated. The team reduced the labor and cost of assembly by making all parts modular and easily accessible. Innovations By employing a systematic design approach, Team Daedalus could readily incorporate selective innovations at the conceptual design phase of the project. The team learned from earlier designs that simplicity and reliability were critical features of a successful design. Team Daedalus also learned that making the vehicle light and compact facilitated transport and testing. These features are especially important for the Disney venue where vehicles may need to be carried or carted from the indoor conference center to the competition course. The team also learned that the user interface and accessibility of key components were important attributes of successful prior designs. While these features individually are not unique to Daedalus, taken in combination, they represent a level of refinement not seen in previous vehicles. Figure 3 shows a front view of the vehicle with the front Lexan enclosure open, allowing access to the electrical box and batteries. Figure 3. Daedalus accessibility Another innovative feature of Daedalus is wireless Ethernet communication with the on-board computer through PC Anywhere. This provides two significant advantages. First, it allows the operator or programmer to monitor the running vehicle remotely and in real-time. This makes testing easier, and debugging the vehicle code more efficient. The second advantage of wireless Ethernet is that it eliminates the space, weight and power consumption associated with an on-board monitor and keyboard, which in turn reduces the size of the battery pack for a given run time. This helps to make Daedalus a lightweight and compact vehicle. In larger production runs of a commercial vehicle, this configuration would also allow a customer to program and control multiple vehicles from a single base station. This would in turn lower the production cost of 4

5 each vehicle. In retrospect this approach seems obvious. Why include an operator interface on a vehicle that is intended to run without an operator? Figure 4 illustrates testing and debugging performed from a remote base station using a wireless Ethernet Ad Hoc network. Another small, but helpful, innovation is the use of a four-bar linkage to guide and support the rear weather-resistant cover of the vehicle. Most of the vehicles entered in previous competitions used a latch or bolt-on cover that needed to be completely removed from the vehicle to allow access Figure 4. Debugging from a remote base station to the interior components. The linkage-guided cover on Daedalus rests above the vehicle in the open position and closes the hood securely during operation. Previous competitors cited space limitations as an important issue in critical situations, such as the work areas provided at competition or when the vehicle was in the queue waiting to compete. Figure 5 illustrates the operation of the linkage. Figure 5. Four-bar linkage operation illustration An important, but invisible, innovation is the navigation software implemented on Daedalus. This software is a variation of the vector field histogram approach that was used by the Virginia Tech Navigator team last year. Rather than writing separate algorithm code for the Autonomous Challenge and the Navigation Challenge, Daedalus uses the same approach and code for both competitions. A target point is set, either using computer vision to locate a point on the course or by using differential GPS to target the next waypoint. The navigation system then tries to move the vehicle in the general direction of the target point, modifying the path with weighting functions to avoid nearby lines, potholes or obstacles. In the absence of a clearly defined target point, the vehicle tries to move straight without crossing a boundary or hitting an obstacle. 5

6 The navigation code treats lines, potholes and barrels as obstacles to be avoided; hence the vehicle will always be viewing an obstacle during navigation along the course. Frame The vehicle frame is compact at 2 wide x 2 high x 3 long, and is constructed using lightweight aluminum. The frame members are welded together forming a solid, stable platform that eliminates the added weight and complexity of bolted joints for fixed components. Sensitive components are encased in a weatherproof Lexan cover. The front door of the vehicle locks in a closed position and easily swings open to access the electrical box and to replace drained batteries. The rear portion of the cover connected by a four-bar linkage, swings up and away allowing access to the computer, payload, and electronics, and locks in open and closed positions. To eliminate excessive wiring and complexity, electronic components are well-labeled and confined to a central drawer that easily slides fully out of the vehicle. This modular design isolates components, allowing for easy access and maintenance. Two drive motors are mounted under the frame to lower the center of gravity, thus improving the vehicle s stability and mobility. At the base of the vehicle, two beams and a plate support the computer and motors. Eight batteries rest in two levels at the front of the vehicle. With this configuration, the center of gravity is located slightly behind the front drive axis. A center beam, made from aluminum box tubing, provides stability, supports the hood, and provides a connection for the electrical box. The hood of the vehicle is made from durable Lexan, which provides a sturdy, weatherproof protection for sensitive interior components. A four-bar linkage connects the rear hood to the center beam, while a hinge assembly connects the front door. The materials and layout of the beams provide a lightweight, versatile design, which allows for quick and easy access to interior components while providing excellent support. Drive System Daedalus uses two Bodine 24 volt DC, 15 amp, 0.45 hp, brush-type servomotors to power the drive wheels. These motors are attached to the drive wheels via two 90-degree, 33:1 reduction gearheads. Along with a rear caster wheel, this differential drive system provides an excellent balance of mobility and stability. This differential drive system allows for zero-radius turns with minimum wheel slip. The hard drive wheels also contribute to reducing slip, thus reducing error during navigation. This allows the vehicle to navigate accurately using dead reckoning, which is used during the GPS 6

7 Challenge portion of the competition. Aluminum hubs connect the drive shaft and the wheels. The motors include integral optical encoders and fail-safe brakes that stop the vehicle when no power is available. Electrical System Overview: The team s desire to construct a safe, reliable, efficient, and serviceable electrical system drove Daedalus s electrical design. The electrical layout was developed and refined in CAD to reduce wiring time and to provide documentation for future Virginia Tech teams. Power System: Rechargeable 24-volt DeWalt NiCad batteries are used to power all systems on Daedalus. A total of eight batteries are used, three for the motor bus and five for the computer bus. In normal operation, the vehicle operates for 3 hours on a fully charged set of batteries. A hot-swap battery system was implemented for replacement of drained batteries without having to shut down the entire vehicle. In our interviews with students from earlier competitions, this ability was cited as a significant advantage on the day of competition. Teams with hot-swap batteries were able to practice or wait in a queue for their turn to run without extension cords or fear of running low on power. Low loss diodes have been added in series to each battery to prevent back-charging. In addition, metal polypropylene capacitors and fuses are placed at each battery and motor to reduce the effects of current spikes and noise which may damage onboard electrical equipment. Relays placed in series with each battery allow temporary bus disconnection for checking battery voltage levels. This is accomplished using a National Instruments PCI-1654 board to sequentially trigger eight relays and check the voltage of each battery as well continually monitor the motor and computer bus voltages. Labview 6.1 has been used to develop the battery check software interface as shown in Figure 6. This software checks battery voltages on command or at timed intervals and estimates Figure 6: Battery check software remaining battery life. 7

8 Control System: Daedalus uses a closed loop PID (Proportional Integrated Derivative) control system to drive each wheel as shown in figure 7. A Galil DMC-1030 motor controller capable of multi-axis control sends a +/- 10 V analog signal to each of two Advanced Motion Controls amplifiers. The amplifiers in turn send a corresponding current between 0-15 A to the drive motors. An encoder integrated into the motor reads the velocity of the motor and sends it back to the Motor Encoder Amplifier Motor Control Board PC motor control board. Based on the encoder reading, the motor control board sends an updated signal to the Figure 7: Control System amplifiers and the cycle repeats. Electrical Enclosure: To ensure that Daedalus s electronics are easily serviceable and protected, a 17 x 17 x 4 aluminum enclosure houses key electronics as shown in Figure 8. This innovative packaging protects electronic components and allows them to be removed for servicing in less than a minute. In our prototype vehicle, this allowed easy bench-testing and repair of electronic components. In a production vehicle, this arrangement would have the significant advantage of modular replacement of on-board electronics, which would permit rapid re-deployment and off-line diagnostics Figure 8: Electrical housing enclosure and repair. Twelve labeled connectors on the electrical enclosure provide a quick disconnect and clean appearance. These include connections to the motor, laser rangefinder, computer, and power as shown in Figure 9. Figure 9: Rear electrical housing enclosure 8

9 Safety Safety was our most important concern in all aspects of design, fabrication, and operation of Daedalus. Since this topic is so pervasive in the design process, discussion of safety is distributed throughout this report. Especially important at competition are the fail-safe emergency brakes and E-stop interlocks. The vehicle is hardware limited to approximately 5.8 miles per hour. Software is used to limit the top speed to 5 mph to conform to competition rules. As part of the design program at Virginia Tech, safety is also taught and emphasized in all student machine shops, welding shops and laboratory work. All students who performed machine work were machine shop certified through Virginia Tech. Fuses added at each battery and on each power bus protect against power surges in the electrical system. An emergency stop control, located at the rear of the vehicle, opens two relays (one for each motor) disconnecting power to each motor. In addition, the motor brakes are automatically triggered once power is disconnected from the motor, stopping the vehicle immediately. With a range of approximately 30 feet, a remote emergency stop feature is programmed onto a wireless remote joystick. When activated, the software detects that an emergency stop command has been activated and sends zero state power to the motors. Software Data Acquisition, Manipulation, and Interpretation: To command the drive system to follow lanes and avoid obstacles, raw data must be acquired, processed, and fused by software running on a computer onboard Daedalus. Reliable, modular base software, developed by previous Virginia Tech students and tailored to fit the needs of Daedalus called Navigation Manager, is used to perform this task. To acquire a visual, two-dimensional image of the immediate area ahead of the vehicle, a Sony 8mm color camcorder is used. Using a Sick LMS-100 Laser Rangefinder, three-dimensional obstacles are detected that may not be interpreted from camera data. Both sets of data are then fused together into a vector field histogram. Statistics is then used to determine the best vehicle path. During the Navigation Challenge portion of the competition, a differential GPS system combined with dead reckoning provides the means to navigate between waypoints. In combination with the vector field histogram, obstacles preventing a straight path between waypoints are determined and avoided. This section describes how Navigation Manager interprets data from the sensors described above, and based on those interpretations, commands the vehicle to follow painted lanes, avoid obstacles, and seek waypoints. 9

10 Navigation Manager Software: The Daedalus team adapted the modular Navigation Manager software to fit needs set forth at the onset of design. Using Microsoft Visual C++ under Windows 98, complex elements of past navigation schemes are eliminated, such as simultaneous frame grabber processing for two cameras, while fine-tuning and tailoring software parameters to fit Daedalus hardware. Built -in, real-time access to sensor and operational data, available through on-screen displays, aided the team in its customization and testing processes. Figure 10 shows a sample interface of the software, illustrating its autonomous mode operation options. Figure 10. Navigation Manager Interface Lane Following and Obstacle Detection: To navigate between a set of yellow or white painted lines, solid or dashed, Navigation Manager applies image processing to frame grabber-converted digital images acquired by the camera. This processing includes running through algorithms that convert the image to grayscale, blur the image, apply threshold, and then decimate the image. This processing is performed to search for the relative light and dark features in the newly acquired image, while ignoring everything else. The navigation software interprets the brightest pixels as two-dimensional obstacles, such as lanes and potholes and will command the motors to steer to avoid them. If no lanes or potholes are discovered, the vehicle continues to move forward in a straight path. If an entirely bright image is acquired, the vehicle interprets this as a sand trap and moves forward until another lane or obstacle can be distinguished. Image acquisition and processing to recognize a bright white lane in a grassy field is shown in Figure 11. Figure 11. Image Processing To Detect Lanes and Obstacles 10

11 To further improve accuracy in obstacle detection, a Sick LMS-100 laser rangefinder scans for three-dimensional objects in front of the vehicle over a 180-degree range in half-degree increments. This scanning improves the accuracy of navigation decisions and overall safety by detecting obstacles that may be path inhibitive, but not bright in color. When a three-dimensional obstacle is detected, such as a barrel, the rangefinder returns the distance and angle to the obstacle and Navigation Manager can calculate the placement of that object within the vehicle s field of view. Information gathered from the laser range finder and camera is passed to the Navigation Manager software, which then fuses all data and makes informed navigation decisions. The fusing of both camera and laser rangefinder data is accomplished by an algorithm that combines the collected obstacle data into a composite Vector Field Histogram. This polar plot of obstacle density at all angles within the vehicle s field of view allows Navigation Manager to select the best path for Daedalus to traverse. If a path suitable for forward travel is not found, the vehicle pauses due to possible glare or noise. If after a short period of time the glare does not subside, a trap is recognized and escape function called based on the data. A sample of the Vector Field Histogram data is shown in Figure 12. In this screenshot of the vehicle running in autonomous mode, the best path based on fused data is shown in red. Figure 12. Fused Vector Field Histogram Showing Best Path in Red 11

12 Global Positioning System (GPS): Many considerations were given to select a method of determining Daedalus s position for the Navigation Challenge. The two most promising methods incorporated using dead reckoning analysis or a Global Positioning System. Dead reckoning analysis is the method of determining position points used by former vehicles. This method calculates the position of a vehicle by using the velocity and direction data that is used to drive the motors. The accuracy of this method degrades with the occurrence of wheel slip on the driving surface. While a portion of wheel slip occurs during straight path travel, a great magnitude transpires while the vehicle turns. During last year s competition this error was significant as the challenge took place on grass, which provides poor wheel traction. The 10 th Annual IGVC competition will occur on pavement, which has a much higher coefficient of friction. While some wheel slip is inherent on any surface, Team Daedalus expects the magnitude of dead reckoning error due slip on pavement to be low. In past competitions, Global Positioning System (GPS) was utilized in the Navigation Challenge. GPS receives signals from multiple satellites and triangulates the data to determine the location of the vehicle. This year, IGVC rules allow teams to use a Differential Global Positioning System (DGPS). Such a system receives a differential correction signal sent by land beacons to improve data acquired by GPS. These signals are statistically analyzed to provide optimum positioning. By utilizing the advantages of both approaches to determine position, Daedalus is expected to complete all waypoints in the Navigation Challenge. Dead-reckoning and DGPS is combined using a weighted algorithm embedded in the control code. As shown in Figure 13, the estimated error for the dead reckoning analysis is in red, while the DGPS system, in blue, holds a constant sub-meter error. Also shown in Figure 13 in green is the approximate algorithm used to combine the two methods. The Differential Global Positioning System that Daedalus uses is an all-inclusive Trimble AgGPS-132. The algorithm favors dead reckoning analysis at the beginning of travel. As dead reckoning accumulates error, the algorithm will transfer weight to the DGPS unit. 12

13 System Postion Error Dead Reckoning Analysis Differential GPS Combined Algorithm 1 Error Displacement Figure 13: Position error with displacement using different methods of waypoint navigation. Testing The results of several tests on mock Autonomous Challenge and Navigation Challenge at Virginia Tech show favorable results in Daedalus s performance. An obstacle course was set up to include 8 lanes with solid and dashed boundary lines. Obstacles placed within the lanes included a painted pothole, cardboard boxes, a barrel trap, and several construction cones. The vehicle s camera position, angle with the horizontal, and amplifier gains were optimized during preliminary tests. Further observations during testing confirmed the need to move the laser range finder from 2 to an adjustable 2 to 16 above the ground. Once these changes were made, the vehicle successfully traversed 9 out of 10 laps in the mock course. GPS testing has been accomplished in parallel with autonomous mode tests to aid in the development of the aforementioned control algorithm for the Navigation Challenge. Testing has given Team Daedalus an idea of a few problem areas that the team will work to resolve before competition such as the vehicle s handling of a zero-possibility situation. In repeated tests, whenever the vehicle had no best path to follow, the vehicle would stop. The team will continue to optimize the vehicle s many adjustable parameters, and will continue to perform more controlled tests before competition. 13

14 Cost Table 1 lists the estimated retail and the actual academic costs incurred by Team Daedalus in the construction of its autonomous vehicle. Table 1. Estimated Retail and Academic Costs for Daedalus Component Retail Cost Academic Cost Laser Range Finder, Sick Optics $ 5, $ 2, Computer, Industrial Computers $ 2, $ - Motor Controller, Galil $ 2, $ - (2) Drive Motors, Bodine $ 1, $ - Batteries, DeWalt $ $ - (2) Amplifiers $ $ Alumimum Frame $ $ mm camcorder, Sony $ $ Electrical comments $ $ Lexan housing $ $ Battery mounts, DeWalt $ $ - (3) Wheels $ $ Camera Mount $ $ Total $ 13, $ 3, Team Information Nicole Anthony Mechanical Engineering Senior Byron Collins Mechanical Engineering Senior Michael Fleming Mechanical Engineering Senior Chuck Liebal Mechanical Engineering Senior Michelle Nicholas Mechanical Engineering Senior Matt Schmid Mechanical Engineering Senior Total Hours: