National Robotics Initiative 2.0: Ubiquitous Collaborative Robots (NRI-2.0)

In NRI PI meeting on November 30th, 2016, Reid Simmons (a new NSF program officer who is most in robotics) gave us a better picture of the newly announced so-called NRI-2.0. The new proposal due day is February 2, 2017. After the meeting, I have better understanding of the new trend the NSF is steering us to. Here I describe what I understand.

The new program focuses on ubiquitous collaborative robots. The “ubiquitous” became popular with the success of smartphones. Sensors in smartphones are with us all the time. So we have ubiquitous sensors. Now, naturally we want to have ubiquitous actuators. So, in the future, everyone has a house, a car, a computer, a smartphone, and a robot.

In a sense, robots are in different places already. But they haven’t penetrated many domains as deep as in factories. Here and there, we see robots working in hospitals, farms, mines, and even mars. But they are still very rare. When they are working in our homes, they are usually in disguised form, such as washer and dryers, dishwasher, and Roomba. We haven’t seen a fully-functional butler that can help us, especially the elderly and people with disabilities in daily-living activities.

Even factories can use more robots that are safe to be around so that human workers and robots can work together without segregation. We should see humans and robots as teams coordinately on tasks much more efficiently.
So robot right movement and robot live matters. Right-to-work movement.

So, the NSF puts 30 to 45 million dollars every year into the NRI-2.0 program to support 40-70 projects that would make it happen.

Different from the “old” NRI, the NRI-2.0 emphasize on scalability and connectivity.

Scalability include the following problems:
1. how multiple agents (humans and robots) interact and collaborate so they are effective.
2. How to scale it to different settings (environment), generalization problem
3. How to use online information or information from other robots to learn
4. How the hardware and software are scalable. more complicated system, still reliable.

Connectivity involve the following problems:
1 Make human-robot connection (“old” NRI)
2. share learning from other robots,
3. share sensor information
3. share planning
4. With other sensors on smartphone, IoT.

The program also encourages educational curricula toward this direction, and collaboration with industry and government agencies.

Here are some examples. If you are at work, look around. Now think about what would happen if half of your colleagues decide to quite their work to pursuit their dream of becoming artists, and robots are hired to work with you. You will need to coordinate with not only people, but also robots. It is not completely new to you, since you have used computers and some kinds of machines before. A robot is an integration of a computer and a moving machine. After a while, you will learn how to deal with those robots and duck when a robot arm swings to your head. If the robots are smart, then the learning curve is reasonable and you are happy to work with them. If they are dumb, you probably will completely ignore them and try to finish all the work by yourselves.

A smart robot in a work space with human colleagues should understand not only what it should do, but also what its co-workers are doing. It should have a big picture of what is going on, who is doing what, who needs help, and how to help. The robot needs not only have a profound understanding of the tasks, but only “read” other co-workers’ minds and even anticipate what they will think next.

When there are multiple robots and human workers, things become more interesting. There will be communication among the robots. They will share data, perceptions, and learning very efficiently. They can plan and act coordinately. Ubiquitous co-robots would make understanding the situation and read humans much easier than a single co-robot, because together they have much better perception and can amplify their strength.

Work places are different. A group of robots should be able to quickly adapt to a new environment like humans do, without any special programming. Both the software and hardware of the robot should allow them to take on various tasks, collaborate with different people, and survive in diverse environments and conditions.

If not hard to imagine that the advent of ubiquitous co-robots will trigger more positive and negative feelings. Co-existing with robots would raise new ethical and legal issues as well. We would need to anticipate and understand the complexities of the new human-robot society. When the technology is in development, we should steer it to increase wealth and improve life quality for everybody and prevent social inequality.

Toward this goal, the program listed a few important technical aspects that should be addressed:
1. Collaboration:
Collaborate and coordinate among all agents: humans and robots.
Multiple robot should perceive, learn, plan, and act in a distributed fashion.
Robots can get knowledge and learn from its own experience, each other, humans, and Internet.
Robots should share their knowledge and learning.

Interaction:
Communication should be smooth and efficient so that the learning curve is reasonable.
Communication includes voice language and body language.
Robot should be able to understand and predict people’s behaviors.
Robot should be able to read the people’s mental states, such as a person is happy, sad, or angry.
Robot should be able to develop a trust-worthy relationship with humans.

Scalability:
The software and hardware should be general. They can be used in different situations.
They should be scalable such as composable, integratable. The software for a team of two robots should be the same for a team of 200 robots.
Data sharing should be scalable as well.

Physical Embodiment:
Robots should be able to collaborate with humans with physical contact.
They should be safe. Soft robots could be safer. But a soft robot can still chock people.
Robots may need a lot of sensors to accurately monitor their state to ensure safety.

In additional, many other aspects of work could help the goal as well:
An easy to use operation system and programming language would allow more people to program a robot quickly without much training. The same thing applies to hardware. We want to lower the bar for people to make and program useful robots.
Shared real robotic platform, software, and data would be very useful for testing ideas, comparing results, and evaluation.

There hasn’t been lacking of speculations on what the world and society would looks like when we are surrounded by robots. Many of them are cheerful, but some of them are scary. But there is a lacking of systematic and scientific studies on this topic. Hopefully, this NSF program will support some project to provide convincing answers on what the social impact would be and how to promote positive impacts and prevent dangerous consequences.

The program will find two types of research projects. One is foundational and the other is integrative. A foundational project would focus on one or two technologies that are important to the development of Co-robot teams. The funding for each project would range from $350,000 – $750,000 for three years. An integrative project tackles a long-term problem by integrating a set of technologies and applying them on real platforms in real environments. The funding for each project would range from half millions USD to 1.5 Millions USD for four years.

I can see the evolving of the NRI 2.0 the old NRI. With the surge of autonomous vehicles (AVs), the co-robot relationship between a driver and an AV and its challenges has been identified, studied, and researched in industry for several years. NSF would want to stay ahead of the curve. For example, a road would be occupied by many AVs and many regular vehicles driven by ordinary people. How the AVs share perception capabilities, data, planning, action, and communication with human drivers could dramatically improve the safety of the traffic system. I am not a good fun of focusing too much on distributed robotics. But, I can see why a group of robots would be a magnitude more powerful than a single robot in co-robot scenarios.