Exclusive interview with Claude Florin, CEO of Fastree3D, on helping robots finally see the light just that little bit better than they did before
How do robots see the world? Until now, most of them have had to make do with conventional digital cameras for eyes. In technological terms, these cameras are much like those available to consumers in the shops and, increasingly these days, in their smartphones. As clever as they are, and as high quality as the images turn out to be, these cameras only capture the image as a two-dimensional arrangement of pixels.
This means that a robot using such cameras would not able to perceive the three-dimensional space its “eyes” are looking at. This problem of perception – of perceiving 3D space as 2D space – is solved, or at least tackled, at the coding stage.
To program the robot to translate the flat image into a three-dimensional space, and infer such things as depth, distance and geospatial location, is really quite difficult if approached from a purely computer programming – as in, coding – standpoint.
The better solution would be to use cameras that send 3D images of what the robot is looking directly into its brain, or its artificial intelligence engine. This would eliminate the need to write probably thousands of lines of code which may or may not work in a given environment.
With such 3D cameras and vision, robots can probably observe and move within their surroundings with much more efficiency and accuracy, using their constantly-updating geospatially accurate model of the world it finds itself in.
The technology to provide robots with 3D vision has been available for quite some time, but only recently has the technology developed to a point where it can be made available at a reasonable price.
One of the companies at the forefront of the innovations that have led to 3D vision technologies being cost-effective enough to use in consumer electronic products is Fastree3D, a Switzerland-based company that grew out of the Bern University of Applied Sciences, the Delft University of Technology, and the Ecole polytechnique fédérale de Lausanne (EPFL).
Armed with more than 10 patents, Fastree3D was recently voted one of Switzerland’s top 25 startups in all categories. And in a more relevant category, photonics, it was considered one of the top 15 startups.
A number of other plaudits have been bestowed on the company and there are high hopes that its visionary technology – pun intended, droll as it is – will be accepted widely and lead to commercial success. The critical ingredient in its recipe for success is something Fastree3D calls its “time-of-flight” technology, or ToF, which the company says precisely resolves the distance based on the speed of light. Sounds clever, and surely it must be if it’s based on the understanding and calculation of the speed of light.
Claude Florin, chief executive officer of Fastree3D, is confident the company has something to offer the market. In an interview with Robotics and Automation News, he says: “Fastree3D Time-of-Flight imager brings unprecedented sensing price and performance for mid-range 3D imaging and fast motion applications.
“The system integrates CMOS single photon detectors (SPAD), laser array illuminators, SoC and FPGA with a developer’s kit. The innovation results from leading research, nine patents, and industrial validation.
“Fastree3D is a fabless semiconductor company that delivers products, services and licensing rights to the robotics and automotive industries. Our goal is to establish our SPAD based on ToF technology as the standard for volume applications in the market.
“During the first five years we will demonstrate the feasibility with a capital-efficient joint-development strategy with partners. After those five years, we will favor either an acquisition or evolve to long-term intellectual property (IP)-licensing business model in which we neither manufacture nor sell semiconductors incorporating our technology.”
The applications for 3D imagers are many and varied. The one that immediately springs to mind is cars, which are increasingly being equipped with safety features such as autonomous emergency braking (AEB), and collision detection systems. Fastree3D itself emphasises technology that will help cars identify and avoid pedestrians, an area that the European car safety group – Euro New Car Assessment Programme (Euro NCAP) – is also looking into, and has already conducted a number of tests.
Euro NCAP’s secretary general, Dr Michiel van Ratingen, says: “These new tests are the first in the world to assess highly automated vehicle features and driver assistance systems from the pedestrian’s perspective.
“Many new cars now offer some form of AEB system that can help prevent car-to-car collisions, but only some are also able to detect pedestrians. By checking the results on Euro NCAP’s website, consumers will be able to verify manufacturers’ safety claims and choose the right AEB option.”
Innovative as it is, Fastree3D does have competitors in this burgeoning business sector. Florin, who studied at the Massachusetts Institute of Technology (MIT) as well as EPFL, says Fastree3D’s competitors fall into two main groups:
- 3D camera manufacturers. 3D camera manufacturers mostly integrating indirect ToF. A few implemented high-end direct time-of-flight for research applications.
- Specialised photo sensor suppliers. These companies offer photo sensors and engineering services similar to Fastree3D. However the majority operates in industries such as telecoms or scientific instruments. Most of them, deliver more expensive technologies (single cell or small array APD, expensive InGaAs, or only indirect ToF modulated light approaches).
Critical competitive advantages.
Florin says: “The Fastree3D implementation integrates a CMOS detector and a semiconductor illuminator, based on the detection principle of Time-Correlated Single-Photon Counting (TCSPC), a form of direct Time-of-Flight (ToF).
“Other commonly used terminologies for ToF solutions are Lidar (light detection and ranging) and ladar (laser radar).
“The other optical measurement technologies offer different ranges and resolutions and are well known: interferometry, triangulation, video analysis (inter-frame, plenoptic, and so on), and indirect ToF (illuminating with modulated light and deriving time from the phase shift of reflected light).”
Florin views the technical and performance benefits of Fastree3D’s system as being:
Higher reliability. We will provide unambiguous measurements, insensitive to background light and color contrast. Customers benefit from meaningful, actionable data for navigation out of complex environments.
Better sensitivity. Compared with competing active illumination cameras, FastSight requires less emitted optical power to allow imaging farther away (>30m) and at higher frame rate (<2ms/image or >500fps at <10m).
In terms of the market for Fastree3D’s technology, Florin says the customers are mainly from outside Switzerland, despite their being many Swiss robotics companies –ABB being one particularly famous example.
He adds that the main business sectors Fastree3D is targeting include “advanced machine vision, also known as ‘industrial vision’ or ‘vision guided robotics’, focused on computer vision in the perspective of industrial manufacturing processes like quality control, automated manufacturing robots, autonomous vehicle guidance. Machine vision system typically performs ‘sensor fusion’ from of multiple cameras capturing 1D, 2D, 2D ½ or 3D image data, and one or more control system processing such data to extract shapes, distances, speed”.
As for future plans, Florin says: “We intend to operate a fabless semi-conductor business with outsourced camera manufacturing. Our core control will be the subsystem SoC [system-on-chip] sensor design – including masks, layouts and embedded software algorithms for readout.
“We can deliver subsystems or cross-license the SPAD technology, on a nonexclusive basis, to image sensor semiconductor manufacturers or industrial vision camera integrators.
“Our technology will be used for security cameras as well as Advanced Driving Assistance Systems for automotive industry. Our future SoC platform (SDK) will be as flexible as possible in order to be integrated in many applications as possible. The resolution of the SoC is targeted to have a QQVGA [quarter quarter VGA] resolution.”
Only just into its third year of operation, Fastree3D has a team of seven key personnel, two of whom are professors and one a doctor.
Florin says the key trends of interest to the company are machine vision in general, automotive and consumer markets – smartphones and tablets.
As for its future prospects, Florin has embraced the world of business. He says: “We are a fully commercial company as Fastree3D was born in 2013 as a spin-off from EPFL and TU-Delft. Bern University is only a research partner on a special research project.
“It is clear IP development is ongoing. From 2014 onwards, we really saw an increasing demand for direct ToF. Before 2014, there were a lot of industry companies thinking this would remain exclusively for the academic domain.”
Sounds like the accelerated development in robotics and related technologies, as well as its acceptance in the wider market, is something Florin did not expect to happen so soon. But the evidence is there for everyone to see if they have the means to do so – robotics and automation technology is gradually taking over the world. It’s just that most people hadn’t noticed until now.