840: Delicate Viticultural Robotics
09:54 Share

This is episode number 840 on Viticultural Robotics.

Welcome back to the Super Data Science Podcast. I'm your host, John Krohn. Let's start off this episode with a couple recent reviews of the show. The first one's from Parker Moesta, who's a data scientist for Domino's Pizza based in Michigan. Parker says that he's a huge fan of the Super Data Science Podcast. He says it was really transformative for him while he was in grad school. Awesome. Glad to hear that it was helpful for you then, and I hope it continues to be helpful for you through your career, Parker.

Our second review here comes from Dr. Paolo Vinicius Borges, who's head of AI R&D at a firm called Orica in Queensland, Australia. Dr. Borges says, I commute to work twice a week, totaling six hours of travel, and the Super Data Science podcast makes me wish I had more commute time. I have to re-listen to episodes to fill time. That's cool. Well, we're definitely happy for you to be making the most of the show in that long commute.

delighted to be part of your journey there, your literal journey in this case.

Thanks to everyone for all the recent ratings and feedback on Apple Podcasts, Spotify, and all the other podcasting platforms out there, as well as for likes and comments on our YouTube videos. Apple Podcast Reviews seem to be especially helpful to us because they allow you to leave written feedback if you want to, and I keep a close eye on those, so if you leave an Apple Podcast Review, I'll be sure to read it on air like I did these reviews today.

All right, into the meat of our episode. I've been excited all year this year about the potential for AI to revolutionize agricultural robotics and help us feed the planet with high quality nutrition. So I'm jazzed today.

to be digging into an innovative application of computer vision and robotics in agriculture, specifically viticulture. That's viticulture, V-I-T-I culture, which is a word that I candidly did not know before doing research for the show, but it's the delicate cultivation of super expensive grapes.

for making wine. And yeah, wine may not provide the world with high-quality nutrition, but the same technologies developed for delicate wine grapes will be transferable to other plants as well.

And in the meantime, it will help us harvest some delicious wine. So to set a bit of context, while agricultural automation has made significant strides in recent years, thanks to tech like GPS guided harvesting systems and mechanical fruit collection, the automation of grape harvesting presents unique technical challenges.

And this is because the sensitivity of wine grapes, particularly those destined for premium wines, which are valued at about $6,000 per ton, these premium wine grapes have traditionally precluded mechanical intervention because of how sensitive these grapes are. You know, it's difficult to pull them off of the vine delicately enough with human fingers or get robotic ones.

The innovative news now, announced a few weeks ago, is a collaborative project between Queen Mary University of London and a startup called Xtend Robotics that are together working to overcome these grape harvesting challenges through an integration of advanced sensing systems and precise robotic control.

The system under development combines two key technical components: spectroscopic analysis for ripeness assessment and pressure-sensitive mechanical manipulation for collecting those very sensitive grapes.

So let's talk about that spectroscopic system first. The spectroscopic system employs a technique called transmitted light analysis to determine grape composition. This transmitted light analysis technique, while commonly used in laboratory settings, presents significant challenges in the conditions of real-world fields. The system measures wavelength absorption patterns to assess sugar content.

So to assess sugar content, this is a critical indicator of grape ripeness. However, the spectral data of light contain an abundance of information that requires sophisticated filtering to distill meaningful signal about sugar content from all the light noise.

To address this data complexity, we've got machine learning. The researchers implemented a machine learning model specifically trained to isolate relevant spectral signatures from environmental noise. This AI system focuses on identifying key wavelength patterns associated with grape ripeness while disregarding irrelevant data from the surrounding environment.

The current implementation relies on human-in-the-loop control through a virtual reality interface, specifically using Meta's Quest 3 headset.

This temporary solution serves two purposes, and this could be useful to you in tons of different applications. So think about this idea here in the kinds of data problems that you're facing or the kinds of AI model building problems that you're facing. This temporary solution allows for precise control during the development phase while simultaneously generating valuable training data for future autonomous operations. So think about that. While you

don't have enough data to start with in this robotics situation in order to be able to pick the appropriate grapes. You got a virtual headset on and you control the robot, which allows you to do the work, get the work done, and that precise control provides training data for future autonomous operations. So you're creating training data and doing the work of harvesting grapes at the same time.

And this brings us to a fundamental challenge in robotics development, indeed a lot of AI model development, that's the data acquisition bottleneck. In robotics specifically, the field of robotics faces a circular problem. Autonomous systems require extensive training data, but gathering these data typically requires functional autonomous systems.

The current VR-controlled system provides a practical solution to this problem in the case of this viticultural application by enabling human operators to generate high-quality training data through remote operation. The project team has actually deployed this tech into the field, into the real world, in a sparkling wine vineyard called Saffron Grange in the United Kingdom.

And yeah, so that vineyard, that sparkling wine vineyard has implemented an innovative approach to system development and validation. They've designed a controlled test area for robot operation and are providing comprehensive training data in the form of leaf, grape, and juice samples for AI system refinement. In addition to the standalone testing

capabilities of these robots that are being developed for viticulture. A further interesting aspect of this implementation plan involves leveraging global time zones for continuous operation. So the project envisions using skilled operators in Australia to control the robots during UK nighttime hours, effectively enabling 24 hour harvesting operations. This approach not only maximizes equipment use, but also addresses labor shortage issues during critical harvest periods.

In time, as more and more data, as more and more training data are gathered, a labor shortage issue could be eliminated entirely through completely autonomous robotics. Indeed, the project's scope extends beyond immediate harvesting applications. The research team is developing a static monitoring system using an array of spectroscopic sensors, those same kinds of sensors we were talking about earlier in the episode, but these ones, this static constant monitoring system will be used for continuous vineyard monitoring.

This system would enable real-time tracking of ripeness progression, disease detection, and optimal harvest timing determination, essentially creating an automated precision agriculture platform that may not depend on intensive human labor at all.

From a data science perspective, this project exemplifies several key challenges in applied machine learning, including things like real-time signal processing, environmental noise reduction, and the integration of human expertise in training data generation. The success of this system could provide invaluable insights for similar applications in precision agriculture and automated harvesting of other delicate crops.

making it a step on the way to more and more scaling up of agricultural robotics to provide high quality nutrition to everyone on the planet. All right.

That's it for today's episode. If you enjoyed it or know someone who might, consider sharing this episode with them. Leave a review of the show on your favorite podcasting platform. Tag me in a LinkedIn or Twitter post with your thoughts. And if you aren't already, of course, be sure to subscribe to the show. Most importantly, however, I hope you'll just keep on listening.

I'm so grateful to have you listening. Until next time, keep on rocketing out there and I'm looking forward to enjoying another round of the Super Data Science Podcast with you very soon.