The center wheel pictured above is an unpowered wheel (no motors attached) that rolls along with the robot, and turns the attached encoder. This is usually called an encoder wheel (or something similar). To make sure that the encoder wheel is always touching the ground even when the robot is crossing over on uneven surface (the platforms this season or the pipes last season) there are rubber bands that pull the encoder wheel down. This is similar to how a suspension works in vehicles.
You might be more familiar with teams attaching their encoders directly to the powered wheels and can get sensor values like that, however the issue with attaching them directly to the powered wheels is that wheels can slip and throw off the reading of the encoder. This can be a serious issue when the robot needs to autonomously cross over uneven surfaces such as those that I mentioned earlier.
Take for example robot A has the encoder attached directly to the powered wheels, and robot B has an encoder wheel. Both robots are very light and their wheels skid when they first accelerate. Their task is to park on the alliance platform during autonomous. As both robots drive up the platform, their wheels slip as they try to get enough traction to get on to the platform completely. They have been programmed to drive 1000 encoder counts and stop. In this case the physical robots are not moving, even though the wheels are turning, robot A will think that it has reached the target sensor value and stop because that's how much the wheels have turned, but robot B will continue to drive because the encoder wheel measures how far the physical robot has moved and the robot has not physically moved the 1000 encoder counts yet.
Even on flat surfaces, robot B will drive more accurate distances, even with wheel slip, because the encoder wheel measures how far the robot has physically moved unlike robot A that measures how much the wheels have turned.
The trade off with encoder wheels is that they take up more space and can be a bit harder to design and build. That is why many teams stick to attaching encoders like robot A and might use fusion of other sensors and more advanced programming to improve their accuracy.
Hopefully that makes sense :)
Also to team 240P: Great early season robot, can't wait for Medium Quality!