Do not fear them robots

This year 200 million viewers cheered to to the dance of three robots to Michael Jackson`s Thriller and Beyonce`s Single Ladies during the Eurovision Song Contest in Stockholm.

There is no doubt. Robots are hot.

In the past few years we have seen them entering new areas of life, such as logistics or agriculture. While some of that has been greeted with excitement, there is a lot of skepticism and fear associated with the advent of Advanced Robotics (I will go into a definition in a second).

This post is an attempt to provide an accurate representation of the current state of Advanced Robotics and how the space is likely to develop. As a venture capitalist I usually look at tech trends through the prism of young, fast-growing startups. Often they are a good representation of what is happening in the industry in general. If you are an investor, like myself, or a strategist interested to understand how robotics will influence your company you will likely benefit from reading this post.

Most people have never seen a robot at work and fewer even have had their limitations and capabilities explained to them. So where are robots working today? What companies are active in the space? A few months ago I started to put startups I met into categories to better understand the technology status quo and some white spaces to explore. At the end a taxonomy came out, similar to what Shivon Zilis did to Machine Learning.

Signs of Robotic Revolution

CB Insights, a research company, reports $1.3bn in Venture Capital going into robotics in 2015, making it one of the hottest tech areas. At Hannover Messe, the world`s largest industrial equipment trade fair this April, collaborative robots stole the show. A new robot was launched, with a price tag of €10,000. Called FRANKA and designed by top-notch German engineers, it will likely open up new application areas for robots. Four weeks ago Financial Times posted a series of daily articles on robotics.

Most importantly, the industry is game. Mercedes, BMW and Foxconn announced that they will integrate collaborative robots in their manufacturing. On March 8th Google showed how it applies artificial intelligence software to enable robots to pick unknown objects. IDC, a global research group, projects the robot market to grow at least 17% per annum, to $135bn by 2019. Given the recent advancements in hardware mentioned above, this is likely an underestimation.

What are Advanced Robots?

The word “robot” comes from a Czech play by Karel Čapek from the 1920s. In that play robots were artificial humans used as slave labor in a factory. We started using robots in factories on a large scale only 60 years later. The first application was welding for the automotive industry and the price per unit back then — $0.5m.

These traditional industrial robots are limited to a few axes and cannot move freely. They perform one task along a predefined trajectory. If a task has to be even slightly altered, a specialist would need to take the robot off-line and change the code. This is expensive.

Advanced Robots are supposed to be a bit more like us humans — in terms of their interaction with the environment and the ability to continuously improve their decision making based on external feedback. My colleague summarizes the qualities of Advanced Robots in three simple words: intelligent, universal and flexible. This makes perfect sense.

Intelligent — able to learn or understand things or to deal with new or difficult situations (Merriam-Webster)

e.g. a Roomba vacuum cleaner learns to dodge a chair blocking its way after it has bumped into it a few times.

Universal — existing or true at all times or in all places (Merriam-Webster)

e.g. a Universal Robots arm can be used one day to do pick-and-place tasks and another day to tend a machine.

Flexible — easily changed : able to change or to do different things (Merriam-Webster)

e.g. an ABB Yumi can learn a new task based on a human moving the robotic arm in a certain fashion (teach-in).

Why now (not 5 years ago)?

Advanced Robotics has moved forward so much because of improvements in the following three areas over the past 10 years:

1. Software development: machine learning (esp. deep learning and reinforcement learning) and computer vision. Many of these revolutionary methods ripened up over the past four years. In 2012 Alex Krizhevsky, Ilya Sutskever and Geff Hinton won ImageNet competition with deep learning, which moved computer vision to the next level. It proved that computers can understand and learn from visual input on their own. Just a few months ago Alpha Go computer program beat the best human Go player. This means computers can solve complex, strategic tasks.

2. Hardware development: lightweight robotics and 3D cameras

3. Ubiquitous connectivity & cloud computing

The basic idea behind deep and reinforcement learning is to mimic the activities of neurons in a brain during the learning process through a computer system. The very first algorithms based on the method were developed in 1960s. In the past few years they finally reached commercial maturity due to improvement in algorithms and models, and computing power. Learning is to a large extent a result of pattern recognition. A computer system recognizing patterns in digital representation of sounds, images etc. becomes more and more intelligent. Facebook and Google use it for object recognition in pictures. Robots will use it to learn about and interact with the environment around them (e.g. learning how to grasp objects by trying different strategies and remembering the successful ones).

Ubiquitous connectivity has brought exponential growth to a number of sectors, e.g. retail. For robots being connected to other robotic systems it means that they can instantaneously share their experiences with their peers. This results in learning in exponential fashion. Lets say BMW launches 100 networked robots in its facility, each performs a different task. By sharing their experience, every robot can perform all 100 tasks and each of them is interchangeable. Fanuc, a Japanese giant in industrial robots, announced its plan to connect 400,000 installed machines by the end of 2016.

Last but not least —two significant improvements in hardware have been major turning points for robotics. First was the launch of lightweight robots (LWRs) on the market in early 2000s and continuous price reduction ever since. LWRs find their way into many new applications. Their low weight and torque sensors in each joint make them safer to humans, as they can react to unexpected impacts. They can be placed in our vicinity. The lowest cost LWRs are around €10,000. It is a 20x cost reduction compared to the systems from 2008.

To recognize and locate objects robots need at least two 2D or one 3D camera. The launch of Microsoft Kinect revolutionized the robotics market. This 3D camera was meant for gamers (XBox), but was quickly picked up by the industry, because of its unbeatable price ($150). Over 10x cheaper than other alternatives on the market available at that time. In addition to that specialized chips and GPUs, geared towards robotics applications, have been launched by e.g. Qualcomm and Nvidia.

Many of the developments described above have taken place in the past two years. That is why the time for intelligent robots is now, not 2014 and not in 2020.

What companies are out there?

The taxonomy is not meant to be exhaustive and there are many more companies to be added to the selection. However, the idea was to provide a general understanding of the technologies and applications that have emerged in the sector in the past 5-10 years.

To limit the scope of this research I excluded drones and autonomous vehicles from the overview, even though they fall under the definition of “Advanced Robotics” (intelligent, flexible and universal).

Here are some visible trends:

1. Over the past years robotic companies have increasingly focused on applications, rather than fundamental technology development. The same has happened with applications of machine learning in enterprise software solutions.
2. Logistics in warehouses and medical (esp. surgical) robots have attracted a substantial number of newcomers. Around 80% of tasks in a warehouse relate to transporting items around. Automating these tasks can significantly increase the throughput of logistics centers. Likewise, using surgical robots improves the efficiency of hospitals. They enable minimally invasive surgical procedures, which result in shortened recovery time and fewer infections.
3. Object recognition, picking and robot teach-in (training robots by leading their arms or by demonstration) have been significantly advanced in the past few years bringing new opportunities in navigation, object handling and performing a variety of new tasks.
4. USA, Germany and Japan are unsurprisingly the global robotics superpowers. China is for sure as active, but Chinese companies often do not market themselves internationally. Germany has brought a lot of development in robotic hardware and USA — in applications.

Reflections on the space

Based on my conversations with entrepreneurs and investors active in the robotics sector I put together some observations how the sector is developing.

Full stack solutions for shorter time to market

Commercializing in the area of robotics is not easy. It is dominated by a number of large incumbents (e.g. ABB, Bosch, Fanuc, KUKA) who own the end-customer. If you develop a new hardware component or an innovative software you are down the food chain and rely on getting a cut from the big guys. To strengthen your negotiating position try to build a strong IP position or highly specialized know-how (e.g. Primesense vs. Apple).

Because of this commercialization barriers some investors (e.g. Shasta Ventures) invest only in full-stack solutions — i.e. companies that own the entire value chain and supply directly to the end customer. Fetch Robotics is a good example of that. Another company Brain Corp., which until a few months ago only built software, recently launched an intelligent cleaning robot to try to cash in the value created through their software. Their patents cover mostly robotic learning.

With the higher reward comes a higher risk. Delivering turn key solutions to the end customer, means mastering hardware design, inventory management, warranties and customer support. But there seems to be more appetite from VCs for this kind of risk than the risk of being strong armed by the large corporates.

I expect this trend to continue. Visibly more companies, especially in the USA, think in terms of applications.

Software is eating the world

The advantages of putting robotic systems online are significant and similar in nature to the developments in the area of mobile phones 10 years ago, albeit on a smaller scale. Connected systems allow for experience exchange between users, de-centralized software development (apps) and hardware agnosticism (since hardware can be easily replaced). FRANKA aims to create a library of applications for its hardware platform and encourages external developers to expand on it. I believe we will see more of that. Companies leveraging on the connectivity of robotic systems are going to be very interesting.

From perception to robotic control

The progress in robotics until now has been mostly focused on the perception component. Over the next years we will see more and more application of deep learning methods to do motion planning and control for robots. Since these systems operate in the physical world with humans, safety must not be compromised and failure will not be accepted. Interesting research is being done in human movement prediction at a number of universities worldwide. Soon we will see this research being commercialized.

Supervised autonomy vs. full autonomy

For some robot manufacturers, e.g. iRobot, Savioke or Fetch Robotics, making fully autonomous systems has been the top priority. I think that in many cases it does not make economic sense to exclude a human from the equation. Often human guidance makes a robotic system much more efficient. E.g. teaching in gripping strategies is more efficient than letting the robot figure it out itself. Tele-surgery, autonomous vehicles and most of the drones are semi-supervised autonomous systems.

Companies that succeed in figuring out the most optimal hybrid man-robot duty cycle will likely win. I believe that except for highly controlled environments (usually in manufacturing) creating fully autonomous systems will not be very economical.

I am bullish about robots bringing automation to areas where they can be more exact than humans or perform very repeatable tasks, which humans beings are not destined to do. I expect to see more of them in manufacturing, logistics and medical procedures.

However, I do not think that we are anywhere close to having robotic helpers in our households, as presented in some recent popular science books like The Industries of the Future by Alec Ross. Despite the fact that 60% of people would accept robots to take care of them when they grow old, I don’t expect robots to dominate the personal care or household service space any time soon. The tasks are too complex and varied, and the empathy factor too important.

Given the momentum the space has today, it is likely to continue growing aggressively in the next three years. As people get used to seeing robots enter some industrial applications to assist humans, they will likely be less skeptical about their presence outside the factory gates.

***

I hope that this overview is valuable to people relatively new to the space, and of help to those informed about it, but looking to systematize their knowledge. This is meant as a living document, so please contact me if you think that I missed an important category or a company. Thank you to: George Babu, Vaibhav Unhelkar, Justin Bayer, Marius Treitz, Marta Grzechnik, Iwona Cymerman, Lylan Masterman, Jonathan Hakakian, Matt Otterstatter and Daniel Waterhouse for your valuable feedback.