As far as I understand the joint encoders output the angle and the wheel encoders output speed. Is that accurate? In either case, how do the differences affect the kinematic calculations, as in path planning, trajectory tracking and motion control?
Encoders used in wheels and manipulator joints are essentially the same. It is the post-processing pipeline that changes their usage and significance. These encoders basically sense the rotational motion.
All optical encoders or magnetic encoders or hall effect encoders record the amount of rotation that occurred in a fixed time. Generally, a magnetic or hall effect sensor has a resolution of somewhere around 12 bits (2^12 = 4096 ticks per revolution = ~0.087 degrees) or higher.
Encoders used for join angle measurement sum the number of ticks obtained per in a definite period of time to measure the motion of the joint. This is added to the existing multi-turn count to express the absolute value of joint angular position.
However, for wheel encoders, they obtain the number of ticks and then divide it by the amount of time passed to express the same result as joint angular velocity.
Sometimes, depending upon the sensitivity and desired accuracy needed in the task, different kinds of encoders may be used. Because robot wheels are expected to rotate faster and continuously gaining multi-turns, one could expect encoders with somewhere around 10-12 bit resolution for them while given that a manipulator arm needs more accuracy and precision, they could have resolutions as high as 16 bits.