Dyson cleans up robotics with Vicon
There are few names that are as synonymous with innovation as Dyson. The company has become a watchword for invention and improvement ever since its founder and chief engineer, James Dyson became frustrated by his vacuum cleaner’s performance and took it apart to invent the world’s first bagless vacuum cleaner. Today, its vacuum cleaners, hand dryers, bladeless fans, heaters and more are sold in over 70 countries with design quality at the heart of its success.
A key area for future Dyson technology is robotics. The company began working with Imperial College’s Prof Andrew Davison in 2005 and in 2014 announced a new £5m robotics lab based at the college. The unveiling of Dyson’s 360 Eye robotic vacuum cleaner is just the start of exploring future robot technology development for the home and beyond, in both the Imperial College facility and in Dyson's R&D laboratory at their Malmesbury headquarters.
Enabling robots to work in the real world, through improved vision and computer processing power is no easy feat. It might sound easy to navigate a house.But every home has different furniture, curtains, flooring - all laid out differently and liable to be moved. High-performing domestic robots therefore need to undertake complex tasks while adapting to a constantly changing environment – something many existing products on the market simply can’t do.
Dyson has overcome this flaw with traditional robotic vacuum cleaners by recruiting the brightest minds in robotics, working with world class universities and creating best in class research, development and testing facilities at its headquarters in Malmesbury.
Chris Smith, Dyson electronics test systems leader, takes up the story: “At Dyson we encourage our engineers to test ideas rather than spending lots of time thinking hypothetically. That means we do lots of testing and validation particularly around robotic behaviours. We need to be able to test for every user interaction we can think of.”
When Dyson began planning for a new advanced robotics laboratory at its Malmesbury headquarters they started looking at new motion capture systems to further increase their understanding of machine behaviours. Much of the motion capture work takes place in the development testing phase to test the implementation of designs and software before products go into home trialling tests.
Vicon technology has given them the opportunity to get real-time and more accurate data as well as gain greater insight into measurements that were not previously obtainable with the original system such as heights, pitches and rolls. The Bonita family of cameras can capture speeds of up to 250 fps, which means Dyson can easily and precisely capture fast moving objects and track multiple items in the laboratory workspace. The robot testing team combines the motion capture technology with optical recording and a daylight simulator (which replaces the role of the sun) to ensure the lab is always at its optimal set up and closely resembles real-world conditions.
The cameras have been vital in helping the team test the robot behaviours. It’s used to test overall cleaning patterns such as where the robots go and how they got there. They have also used it for testing and improving navigation systems. Finally it’s also used for tracking control, to ensure the robotic cleaners take the most efficient cleaning paths, and object tracking the robot’s ability to detect and navigate around obstacles in the home.
Mike Aldred, Dyson Robotics lead, said: “Motion capture has allowed us to split and co-develop testing so we can help the robots to better understand ‘where they are’ and behaviours ‘what they do with where they are’. The Vicon technology allowed us to test extremes - it meant we could work out how much tolerance the robots will take before they break. We’ve been able to take them to high speeds to test safety, evaluate reaction times for movements and even look at what happens if they fail at speed.”
“It helps us gauge how the robots will act in different scenarios such as on different types of carpet” says Aldred. “If you’ve got a shagpile carpet rather than a traditional household carpet it means we can really understand exactly how the cleaner sensors would react, and how their movement could affect cleaning ability in order to give us a better sense of the error model. Having real-time data allows us to test, see the results and immediately react to them.”
The results from the testing haven’t thrown up major surprises but when things did go wrong it has helped them confirm that their thinking was on the right lines. For example it helped the team to analyse when slips or traction in carpets were caused by bumps – so the robotic cleaners were able to make a decision on what to do next. It’s helped bring together what the robot thinks it’s doing closer to what it’s actually doing.
Improved accuracy of data has also reduced having to repeat tasks with the number of cameras also allowing Dyson to scale up their test area for more scenarios. Aldred adds “Taking out the need to repeat tests means we have to do a lot less iterations. Previously we conducted about 100,000 tests and now we’ve rapidly reduced the amount of tests meaning we are between 25-100 times quicker depending on the tests we are doing.”