Team 31 CART Update (11/01/19)
Problem and Introduction
Here’s the problem: the University of Houston is getting bigger, which means it needs an effective way to transport people (visitors, people unable to walk, etc.) around campus.
The challenge lies in designing an effective ‘short-distance’ transportation system for visitors and students/faculty unable to walk around campus. A potential solution lies in using golf carts to navigate between points of interest on campus. Golf carts are ideal because they are small enough to drive on sidewalks and paths that are inaccessible to regular vehicles. Further, by making these golf carts autonomous, there is no need for a dedicated driver, so the University can reduce the operating costs of this transportation system. Eliminating the human driver reduces cost while increasing the safety of cart, as more than 90% of crashes are caused by human error (NSC 2019).
Unfortunately, there are no commercially available autonomous golf carts. Despite multiple attempts (e.g. Auro Robotics, which pivoted to autonomous cars), no company has been able to deliver a fully-autonomous golf cart, let alone at a reasonable price. The Cougar Autonomous Robotic Transport (CART) team proposes to modify an existing golf cart (which is called the CART) to be capable of level 3 autonomy as defined by the Society of Automotive Engineers (SAE). This means that the CART will drive itself, but requires that a human driver be present to take over in rare situations where the CART doesn’t know how to proceed. We plan to build on the work of a previous capstone team that purchased a used golf cart and installed actuators (steering, brake, and accelerator) to control this golf cart with a remote control (or a computer).
The challenge lies in designing an effective ‘short-distance’ transportation system for visitors and students/faculty unable to walk around campus. A potential solution lies in using golf carts to navigate between points of interest on campus. Golf carts are ideal because they are small enough to drive on sidewalks and paths that are inaccessible to regular vehicles. Further, by making these golf carts autonomous, there is no need for a dedicated driver, so the University can reduce the operating costs of this transportation system. Eliminating the human driver reduces cost while increasing the safety of cart, as more than 90% of crashes are caused by human error (NSC 2019).
Unfortunately, there are no commercially available autonomous golf carts. Despite multiple attempts (e.g. Auro Robotics, which pivoted to autonomous cars), no company has been able to deliver a fully-autonomous golf cart, let alone at a reasonable price. The Cougar Autonomous Robotic Transport (CART) team proposes to modify an existing golf cart (which is called the CART) to be capable of level 3 autonomy as defined by the Society of Automotive Engineers (SAE). This means that the CART will drive itself, but requires that a human driver be present to take over in rare situations where the CART doesn’t know how to proceed. We plan to build on the work of a previous capstone team that purchased a used golf cart and installed actuators (steering, brake, and accelerator) to control this golf cart with a remote control (or a computer).
Problems and Challenges
From October 18 to November 1, we have been focused on thoroughly assessing the CART as delivered by the previous team. While assessing the CART, we discovered problems with the actuators that made it impossible to fully test the cart. For example, the gear connecting the steering actuator to the steering column was disconnected. Additionally, the braking actuator is incapable of pushing the brake hard enough to stop within 20 feet of a detected obstacle, which is a requirement we have set to improve the safety of the cart.
After partially completing the CART assessment, we worked on determining the scope of the CART project. This is a complex decision, as we needed to ensure that we have the resources (money and manpower/expertise) required to achieve the desired scope. We immediately realized that we needed additional team members to complete the project in one year. After a careful search, we decided to collaborate with two members of the University of Houston Makerspace. These Makerspace members are experienced in both computer vision and ROS. Relying on their expertise, we decided to design a ROS-based control architecture for the CART, which we are currently working on.
As mentioned previously, the major challenge we anticipate for deciding on types of sensors is figuring out which are best for our use case. Once this is done, we will decide on which specific COTS sensors to buy, which involves lots of searching and reaching out to vendors.
We also hope to start fundraising during this time. A discussion of the fundraising plan exists above.
Lastly, during this time, we plan to work on some technical documents such as the technical design review and the technical design presentation PowerPoint. These are important because they help document the project and our decision making process.
 |
| Steering actuator (geared DC motor) |
We have designed and priced out solutions to both of these issues. To solve the gear disconnect problem, we could replace the existing press-fit mechanism (installed by the previous team) with a key/keyway assembly. Alternatively, we could re-press the gear onto the steering column and better secure it somehow (set screw, brazing, etc.). Either way, we hope to make our steering actuator safer and more reliable.
Upon further testing, we realized that the existing braking actuator is well-suited to normal driving, just not emergency braking. Instead of a complete replacement, we need an additional, stronger braking mode. For this, we will install a higher-pressure line from the compressor to the piston. This “emergency braking” mode will, when activated, bring the CART from full speed (10 mph) to a stop within 20ft. Please note that this rate of stopping is NOT comparable to stomping on the brakes of a full-sized car - it is a more gentle stop.
 |
| Brake actuator (pneumatic piston) |
After partially completing the CART assessment, we worked on determining the scope of the CART project. This is a complex decision, as we needed to ensure that we have the resources (money and manpower/expertise) required to achieve the desired scope. We immediately realized that we needed additional team members to complete the project in one year. After a careful search, we decided to collaborate with two members of the University of Houston Makerspace. These Makerspace members are experienced in both computer vision and ROS. Relying on their expertise, we decided to design a ROS-based control architecture for the CART, which we are currently working on.
We plan to combine an Intel RealSense depth camera with a medium-range RADAR unit to provide a reliable perception system These two instruments each essentially create their own maps that we will combine to create a complete picture of the environment in front of the CART.
We intend to drive between points of interest on campus by first creating multiple routes that the CART is capable of autonomously navigating. We will create these routes by specifying waypoints that the CART will navigate to while avoiding obstacles. This is simple to do using the existing ArduPilot-ground framework (and its integration with ROS). The image below shows a screenshot from the ArduPilot software. This software lets us set waypoints on the map and it plans the best route between them for us. Then it commands our autopilot system to drive along the planned route using GPS coordinates from the CART.
 |
| ArduPilot software (screenshot) |
While assessing the CART and determining our project scope, we faced a few challenges that negatively affected the rate of our progress on the project. The immediate challenge our team faces is funding. Our tentative budget is $1100 but our projected maximum budget is $2000. We have raised $500 but we will need additional support to complete this project. So far, our budgeting problem has been on the back burner in favor of assessing the cart, defining our problem, creating a good project plan, and meeting deadlines.
We will need to focus on funding soon so that we can afford the first round of components to begin modifying and adding onto the CART over the winter break. We plan to look into both crowdfunding options such as GoFundMe alongside organizational sponsors like UH Alumni Association and potentially interested companies. We will also explore component discount/donation types of sponsors too.
Another challenge that we have encountered (and must be solved by winter break) is to find a suitable workspace to make modifications/additions to the CART. Currently, the CART is parked in a heavily-trafficked public space that is not suitable for any heavy work to the CART. We are waiting to hear back on a potential workspace in the Energy Research Park (ERP) here on the UH campus that would suit our needs perfectly. Alternative options are workspaces at the UH Katy branch or moving to a self-storage facility nearby.
Aside from the logistical challenges the project poses, we are facing some technical ones as well. The most pressing technical challenge is the selection of our main perception sensor. This decision is very difficult because it is very nuanced - the choice determines not only the capability of the cart to map its surroundings and detect obstacles, but it determines the way we design the entire detection system that processes the incoming information and makes decisions. Luckily, our Makerspace collaborators have some experience working with the sensors being considered and we intend to rely on their expertise in selecting a sensor. We are looking forward to completing our sensor selection soon.
Work plan for Nov 1st to Nov 15th
First, we plan to finalize our perception (and other) sensor selections and conduct trade studies to review COTS components. Next, we will finalize our high level architecture, which is to say, our system diagram (how all the components interact, and how the CART is controlled). This decision is driven by both our sensors and also the specific software we choose to use to command and control the CART.We will need to focus on funding soon so that we can afford the first round of components to begin modifying and adding onto the CART over the winter break. We plan to look into both crowdfunding options such as GoFundMe alongside organizational sponsors like UH Alumni Association and potentially interested companies. We will also explore component discount/donation types of sponsors too.
Another challenge that we have encountered (and must be solved by winter break) is to find a suitable workspace to make modifications/additions to the CART. Currently, the CART is parked in a heavily-trafficked public space that is not suitable for any heavy work to the CART. We are waiting to hear back on a potential workspace in the Energy Research Park (ERP) here on the UH campus that would suit our needs perfectly. Alternative options are workspaces at the UH Katy branch or moving to a self-storage facility nearby.
Aside from the logistical challenges the project poses, we are facing some technical ones as well. The most pressing technical challenge is the selection of our main perception sensor. This decision is very difficult because it is very nuanced - the choice determines not only the capability of the cart to map its surroundings and detect obstacles, but it determines the way we design the entire detection system that processes the incoming information and makes decisions. Luckily, our Makerspace collaborators have some experience working with the sensors being considered and we intend to rely on their expertise in selecting a sensor. We are looking forward to completing our sensor selection soon.
As mentioned previously, the major challenge we anticipate for deciding on types of sensors is figuring out which are best for our use case. Once this is done, we will decide on which specific COTS sensors to buy, which involves lots of searching and reaching out to vendors.
We also hope to start fundraising during this time. A discussion of the fundraising plan exists above.
Lastly, during this time, we plan to work on some technical documents such as the technical design review and the technical design presentation PowerPoint. These are important because they help document the project and our decision making process.
