I'm currently doing some research on collaborative robotics. One area of interest is the type of sensor(s) used in these kind of machines.
I had a look at some robots by FANUC and Universal Robots and I've noticed that they do not come equipped with sensors; they are sold as an add-on.
Is this inherent with collaborative robots? Do customers need to buy sensors as an add-on - which has advantages and disadvantages.
Thanks for your help.
This question seems to be about robots collaborating with people -- not with each other. In general, it might be unsafe for a person to stand near a machine that's capable of exerting harmful levels of force in any given direction. One way to avoid that, as is mentioned in the question, is by using sensors to detect any human-like objects in its path.
However, robots can still be safe even without internal sensors. The WAM arm has configurable torque limits -- read from its own motors --, which can be set low enough such that the arm will disable itself if it runs into something.
You could also do what the PR2 robot does: exert very little force in each of its joints.
Keep in mind that just because the robot senses something, it needs a fair amount of intelligence and mechanical capability to act in a way that will protect a person. In other words, the simple safety designs are just as important as the sophisticated ones.
Very little is inherent to collaborative robots, and this is especially true for sensors.
The term 'collaborative robot' is still new in the industrial world and there is a bit of hype and variation in meaning in what exactly they are talking about.
In general a collaborative robot will be one that expected to do its work with a human in its workspace. There is ISO standardization around this and the focus is on safety.
In the domain of collaboration, there are two main things we want to do in my view.
1) We want to know where the human is relative to the robot so we can operate the robot in a more performance oriented manner when the human is far enough away. This could be as simple as always moving safely and stopping (moving even more safely) when in contact with a human. It could be as complex as tracking human kinematics and moving more 'dangerously' based on the distance between the robot and person and the dynamics between them. Think of the knife-between-fingers scene in Aliens.
2) We want to receive commands from the human so that the robot can collaborate. This might be lead-through teach and play programming, or one might simply sit down at a robot workbench and block off the area you want to use with your arms, like a little kid at the art table establishing their space.
You can see this is a big domain with a lot of possibilities. So when it comes to specific products, you will have to sit down with each vendor to find out what exactly is 'collaborative' about their robot and what package of options is needed to achieve that collaboration. There is no standard of what collaboration means, nor what sensing is used.
One thing to note is that collaboration is completely separate from the actual task of the robot. End tooling, tool changers, fixtures, etc all need to be considered in addition.
ST Robotics have an infra-red sensor system that will stop the robot if you get into the workspace. https://www.youtube.com/watch?v=WqJK1xXTWJQ. You still cant actually share the workspace but can work next to it. ~~ David Sands
The other answers and comments point to the ambiguity about what people mean when they say "collaborative" robot (or "cobot" for short), so I won't belabor that point. In my experience, most people refer to a cobot as one that does not require a safety barrier (e.g. laser barrier or monitored cage) while operating and where humans can be in the robot's workspace without worrying about getting seriously injured or killed.
The mechanism to ensure this safety is often internal force/torque sensors that come with the robot and are not add-ons - they are the very thing that make them safe to be around. For example, FANUC's line of cobots have force sensors in the base to detect contact with the robot and ensure a limit on the force that would be applied on contact. Many robots, such as the KUKA iiwa, UR robots, and Sawyer have joint torque sensors that allow force-limiting.
Collaborative features that are useful but not necessary for ensuring safe co-habitance between the robot and human are typically add-ons that need to be purchased. For example, wrist force/torque sensors are usually sold as add-ons since they are only needed by people that need to perform tasks requiring higher-fidelity force monitoring at the end-effector. The wrist sensor can improve safety, but since it only monitors contacts past the wrist it will not guarantee safe operation. Tactile sensors (e.g. tactile skin) I think would be good for ensuring safety if they cover a large extent of the robot, but they are generally quite expensive and is likely why they are not more popular for this purpose.
As others have mentioned, safety for cobots can come not just from the robot's internal sensors but also external sensors setup in the robot's environment. For example, a laser scanner can be used to detect when a human enters the robot's workspace, and operate at a slower speed instead of completely terminating motion. Vision sensors like RGB and depth cameras can also be used to monitor the robot's workspace for human presence and even do things like human intent prediction.
Mechanical compliance is another mechanism to aid in safety. For example, the Baxter robot uses Series Elastic Actuators that have a spring between the motor and the actuator output that make them deformable if a human was contacted (I'll note Baxter also has sonar sensors in its head to monitor workspace proximity). Some of the FANUC cobots are coated in a soft deformable material to reduce impact force and prevent pinching. An interesting extreme of this idea is soft robots.
Lastly, I'll point you to Robotiq's free e-book on cobots. They keep updating it with the newest iterations of cobots and it's definitely worth a read if you're considering purchasing a cobot.
I only just saw your comment above and would like to reply. Baxter is a special case because as you say it relys on elastic drives which simply deform when in contact with a human (or anything else). Force sensors are not the answer. When the robot makes contact with a human hand (as probably the most likely interaction) the robot stops, right? Wrong. It decelerates. Newtons 3rd law says so. And that requires further distance. If the hand is trapped, say between the end effector and the bench injury is possible. Calling it a collaborative robot doesn't mean a thing. A French university measured the force of a UR onto a force plate and it raised over 1500N of force. If that is your hand injury is inevitable. We are currently perfecting a capacitive sensor that fits on the arm (not a new idea). If the human gets near the robot (or vice versa) within the deceleration distance the robot stops just in time. So if a robot is not inherently safe (and in my view none of them are apart from Baxter) then the best answer is to add an external sensor or workspace monitor.
