A team of researchers at Carnegie Mellon University (CMU) are bringing us one step closer to achieving self-driving all-terrain vehicles (ATVs). The team rode an ATV through various different environments including tall grass, loose gravel, and mud to gather data on how the ATV interacted with these types of off-road environments.
Creating the TartanDrive Dataset
The ATV was driven aggressively at speeds up to 30 miles per hour. It slid through turns, went up and down hills, and got stuck in the mud while gathering important data like video, the speed of each wheel, and the suspension shock travel from seven types of sensors.
After collecting all of the data, it was compiled into a dataset called TartanDrive. It includes about 200,000 real-world interactions, and the team believes it’s the largest real-world, multimodal, off-road driving dataset. The data could later be used to train a self-driving vehicle for off-road navigation.
Wenshan Wang is a project scientist in the Robotics Institute (RI).
“Unlike autonomous street driving, off-road driving is more challenging because you have to understand the dynamics of the terrain in order to drive safely and to drive faster,” said Wang.
There has been some previous work carried out in this area, but it often involved annotated maps that provided labels like mud, grass, vegetation, and water. These labels helped the robot understand the terrain it was navigating, but the problem is that this type of information is often hard to gather. It is also fairly generic information. For example, “mud” could mean an environment that is either drivable or not.
Building Prediction Models
With the multimodal sensor data that the team gathered, they could build prediction models that are superior to the models developed with simple and non dynamic data. By driving the ATV aggressively, it became crucial to understand the dynamics of its performance.
Samuel Triest is a second-year master’s student in robotics and lead author of the research paper.
“The dynamics of these systems tend to get more challenging as you add more speed,” said Triest. “You drive faster, you bounce off more stuff. A lot of the data we were interested in gathering was this more aggressive driving, more challenging slopes and thicker vegetation because that's where some of the simpler rules start breaking down.”
While it is true that most of the research and work surrounding autonomous vehicles is targeted at street driving, the researchers say the first applications will likely be controlled, off-road areas. This allows for less of a risk of collisions.
The team performed all of their tests at a controlled site near Pittsburgh where CMU’s National Robotics Engineering Center tests autonomous off-road vehicles.
The ATV was driven by humans using a drive-by-wire system to control the steering and speed.
“We were forcing the human to go through the same control interface as the robot would,” Wang said. “In that way, the actions the human takes can be used directly as input for how the robot should act.”
The research is set to be presented at the International Conference on Robotics and Automation (ICRA) in Philadelphia.
- DeepFace for Advanced Facial Recognition
- Mohan Giridharadas, Founder & CEO at LeanTaaS – Interview Series
- ContentStudio Review: The Best AI Tool for Social Media? (September 2023)
- Decoding Opportunities and Challenges for LLM Agents in Generative AI
- Tailor Introduces ChatGPT Plugin Enabling Conversational Interface for ERP Operations