Robot Harvesting Machine Is Tip Of The Agri-Tech Iceberg

July 9, 2019 by Pat Whetman 24 Comments

Harvesting delicate fruit and vegetables with robots is hard, and increasingly us humans no longer want to do these jobs. The pressure to find engineering solutions is intense and more and more machines of different shapes and sizes have recently been emerging in an attempt to alleviate the problem. Additionally, each crop is often quite different from one another and so, for example, a strawberry picking machine can not be used for harvesting lettuce.

A team from Cambridge university, UK, recently published the details of their lettuce picking machine, written in a nice easy-to-read style and packed full of useful practical information. Well worth a read!

The machine uses YOLO3 detection and classification networks to get localisation coordinates of the crop and then check if it’s ready for harvest, or diseased. A standard UR10 robotic arm then positions the harvesting mechanism over the lettuce, getting force feedback through the arm joints to detect when it hits the ground. A pneumatically actuated cutting blade then attempts to cut the lettuce at exactly the right height below the lettuce head in order to satisfy the very exacting requirements of the supermarkets.

Rather strangely, the main control hardware is just a standard laptop which handles 2 consumer grade USB cameras with overall combined detection and classification speeds of about 0.212 seconds. The software is ROS (Robot Operating System) with custom nodes written in Python by members of the team.

Although the machine is slow and under-powered, we were very impressed with the fact that it seemed to work quite well. This particular project has been ongoing for several years now and the machine rebuilt 16 times! These types of machines are currently (2019) very much in their infancy and we can expect to see many more attempts at cracking these difficult engineering tasks in the next few years.

We’ve covered some solutions before, including: Weedinator, an autonomous farming ‘bot, MoAgriS, an indoor farming rig, a laser-firing fish-lice remover, an Aussie farming robot, and of course the latest and greatest from FarmBot.

Video after the break:

Continue reading “Robot Harvesting Machine Is Tip Of The Agri-Tech Iceberg” →

Wimbledon 2019: IBM’s Slammtracker AI Technology Heralds The Demise Of The Human Player

July 8, 2019 by Pat Whetman 4 Comments

Whilst we patiently wait for the day that Womble-shaped robots replace human tennis players at Wimbledon, we can admire the IBM powered AI technology that the organisers of the Wimbledon tennis tournament use to enhance the experience for TV and phone viewers.

As can be expected, the technology tracks the ball, analyses player gestures, crowd cheers/booing but can’t yet discern the more subtle player behaviour such as serving an ace or the classic John McEnroe ‘smash your racket on the ground’ stunt. Currently a large number of expert human side kicks are required for recording these facets and manually uploading them into the huge Watson driven analytics system.

Phone apps are possibly the best places to see the results of the IBM Slammtracker system and are perfect for the casual tennis train spotter. It would be interesting to see the intrinsic AI bias at work – whether it can compensate for the greater intensity of the cheer for the more popular celebrities rather than the skill, or fluke shot, of the rank outsider. We also wonder if it will be misogynistic – will it focus on men rather than women in the mixed doubles or the other way round? Will it be racist? Also, when will the umpires be replaced with 100% AI?

Finally, whilst we at Hackaday appreciate the value of sport and exercise and the technology behind the apps, many of us have no time to mindlessly watch a ball go backwards and forwards across our screens, even if it is accompanied by satisfying grunts and the occasional racket-to-ground smash. We’d much rather entertain ourselves with the idea of building the robots that will surely one day make watching human tennis players a thing of the past.

Blisteringly Fast Machine Learning On An Arduino Uno

June 30, 2019 by Pat Whetman 23 Comments

Even though machine learning AKA ‘deep learning’ / ‘artificial intelligence’ has been around for several decades now, it’s only recently that computing power has become fast enough to do anything useful with the science.

However, to fully understand how a neural network (NN) works, [Dimitris Tassopoulos] has stripped the concept down to pretty much the simplest example possible – a 3 input, 1 output network – and run inference on a number of MCUs, including the humble Arduino Uno. Miraculously, the Uno processed the network in an impressively fast prediction time of 114.4 μsec!

Whilst we did not test the code on an MCU, we just happened to have Jupyter Notebook installed so ran the same code on a Raspberry Pi directly from [Dimitris’s] bitbucket repo.

He explains in the project pages that now that the hype about AI has died down a bit that it’s the right time for engineers to get into the nitty-gritty of the theory and start using some of the ‘tools’ such as Keras, which have now matured into something fairly useful.

In part 2 of the project, we get to see the guts of a more complicated NN with 3-inputs, a hidden layer with 32 nodes and 1-output, which runs on an Uno at a much slower speed of 5600 μsec.

This exploration of ML in the embedded world is NOT ‘high level’ research stuff that tends to be inaccessible and hard to understand. We have covered Machine Learning On Tiny Platforms Like Raspberry Pi And Arduino before, but not with such an easy and thoroughly practical example.

Fish Hooks Embedded In Robot Toes Make Them Climb Like Cockroaches

June 28, 2019 by Pat Whetman 5 Comments

Take a dozen or so fish hooks, progressively embed them in plastic with a 3D printer and attach them to the feet of your hexapod and you’ve got a giant cockroach!

A team of researchers at Carnagie Mellon University came up with this ingenious hack which can easily be copied by anybody with a hexpod and a 3D printer. Here you can see the hooks embedded into the ends of a leg. This ‘Microspine technology’ enables their T-RHex robot to climb up walls at a slightly under-whelming 55 degrees, but also grants the ability to cling on severe overhangs.

Our interpretation of these results is that the robot needs to release and place each foot in a much more controlled manner to stop it from falling backwards. But researchers do have plans to help improve on that behavior in the near future.

Sensing and Closed Loop Control: As of now, T-RHex moves with an entirely open-loop, scripted gait. We believe that performance can be improved by adding torque sensing to the leg and tail actuators, which would allow the robot to adapt to large-scale surface irregularities in the wall, detect leg slip before catastrophic detachment,and automatically use the tail to balance during wall climbs.This design path would require a platform overhaul, but offers a promising controls-based solution to the shortcomings of our gait design.

No doubt we will all now want to build cockroaches that will out perform the T-RHex. Embedding fish hooks into plastic is done one at a time. During fabrication, the printer is stopped and a hook is carefully laid down by human hand. The printer is turned on once again and another layer of plastic laid down to fully encapsulate the hook. Repeat again and again!

Your robot would need the aforementioned sensing and closed loop control and also the ‘normal’ array of sensors and cameras to enable autonomy with the ability to assess the terrain ahead. Good luck, and don’t forget to post about your projects (check out Hackaday.io if you need somewhere to do this) and tip us off about it! We’ve seen plenty of, sometimes terrifying, hexapod projects, but watch out that the project budget does not get totally out of control (more to be said about this in the future).

Continue reading “Fish Hooks Embedded In Robot Toes Make Them Climb Like Cockroaches” →