For a given stopping force, the effective mass of the robot will determine the required stopping distance. If we pick a stopping force of 150-newtons, often used as a rule of thumb for collaborative robots, we can compare the required stopping distance at several velocities for a high ratio robot and a low ratio robot. This table shows the stopping distance for two robots. A high ratio Harmonic Drive SCARA and a low ratio belt drive SCARA, for both a 500 gram and a two-kilogram payload. At a speed of 500 millimeters per second, the harmonic drive SCARA it takes 12-13 millimeters to stop while the belt drive SCARA takes one to three millimeters to stop. Recall that the human hand has a compression constant of 75 newtons per millimeter so that you can only compress the human hand two millimeters before reaching the 150-newton pain threshold. Therefore, unless the harmonic drive robot has 10 millimeters or more of foam on the part that strikes a human hand, at 500 millimeters per second, it will likely cause injury while the belt drive robot will not. At a speed of 1,000 millimeters per second, the energy is increased by a factor of 400 percent and the stopping distance for the harmonic drive robot increases to about 50 millimeters. Well, the belt drive stopping distance increases to 5-10 millimeters depending on the payload. Collision force tables, if available, can be used to set speed limits for various types of robots in various motions. In general, low ratio robots will be able to incur safe collisions at higher speeds than high ratio robots. In addition to the inertial forces that contribute to the collision force, we have forces from the robot motors. In many robot controllers, when a collision occurs, the path following error increases quickly and the motor torque quickly ramps up to the maximum possible motor torque before shutting down with an error. However, it is possible to design a control system which limits motor torque in the event of a collision. This can be done by developing a complete dynamic model of the robot which computes in real-time the motor torques necessary to make a commanded motion including all the torques coupled back from the other robot axes. This computation is complex but with today's microprocessors, it can be done very quickly. The precomputed torque from this dynamic model is sent to the controller as a feed-forward torque along with the commanded motion. The controller combines this feedforward torque with feedback torque which is computed based on the path error from the commanded motion. If the dynamic model were perfect, there would be zero path error and the feedback torque would be zero. However, since there are always some unmodeled errors such as friction or tooth cogging in the drive train, there will always be some feedback error torque. Low ratio robots tend to have low friction. In these robots, the feedback error torque can be as low as 10 percent of the total motor torque. It is, therefore, possible to set a maximum limit on the feedback error torque, that is, for example, 25 percent of the maximum motor torque. In a collision, this 25 percent motor torque limit is all that is allowed before a shutdown as opposed to a 100 percent limit without this feature. In this image, you can see an example of this control strategy. The green triangular line is the dynamic feedforward computed torque from the robot model. The blue wiggly line is the feedback torque from the path following error. You can see that the magnitude of the feedback torque is much lower than the feedforward torque. The yellow line, superimposed over the green triangular line, is the total torque output from the controller to the motor which is composed of the feedback torque added to the feedforward torque. When the motor is following a normal commanded trajectory, most of the torque comes from the feedforward model and only a small percentage of torque comes from the feedback model. Therefore, it is possible to set a limit on the feedback torque that is about 25 percent of the total motor torque. A collision will quickly cause the feedback torque to rise up to this limit in which case the robot will stop. This limits the collision force from the robot motors to 25 percent of what this force would otherwise be. Note that this strategy is most helpful for the lower speed collisions. For high-speed collisions, the inertial forces will dominate and factors in the mechanical design of the robot such as reflected inertia are most important. Here's a video of a collaborative robot doing a 600-millimeter pick and place in a few seconds with a one-kilogram payload. An operator can literally place his hand in the pinch point of this robot and it will stop without injuring the operator. Note that this collaborative capability is also very useful when the robot is loading an expensive instrument or test fixture. If the robot bumps into the instrument, it does not damage it in the way of more powerful robot might. Here we see a small conveyor with two collaborative robots attached to it. One robot is a SCARA type robot sitting next to the conveyor and the second robot is a collaborative of Cartesian robot which is mounted directly over the conveyor on standoffs These are both low ratio robots with very low collision forces. The low ratios also make it easy to back drive these robots for the purposes of teaching. Let's put this robot into teaching or free mode in which the gravity load is counterbalanced by the motors and we could move the robot into various positions with just a few grams of force. This makes it extremely fast and easy to teach the robot, as you can see. You don't have to fight with high ratio of gear trains to position the robot accurately. Another benefit of the back probability of these robots is the ease of compliant insertion. If parts or nests have a modest chamfer on them, robots that are back-drivable can easily comply in x, y and Theta to allow the insertion to be completed without the jamming that often occurs with robots with high ratio, very stiff, gear trains. Now, let's start at this line and see these robots in action. I can walk up and easily stop the robot without any large force. In this case, the robot has been programmed to remember what it was trying to do. Even though it was interrupted, it pauses for a moment and then resumes operation and completes its task. Let's look at the pinch point on this SCARA robot. I can place my fingers directly in the pinch point and the robot stops easily without injuring my hand. Then, after a pause, presumes its operation. The ability to bump a person without injuring the person and then recover and continue performing its task allows this class of collaborative robots to be intermixed with people on the same assembly line. These robots are moving at speeds which do not scare people. If they bump the robot, the robot stops and the person is not injured. Then, the robot can recover without the need to restart the whole line from an emergency stop shutdown. The ability to play is friendly, collaborative robots in convenient places in existing lines greatly reduces the time and expense to introduce robots into the simple applications where they perform best. People can continue to perform more complex applications and the robots can do the boring jobs.
