To give a brief answer: BinauralPan is a pan that uses both interaural intensity differences and interaural delays to model how sound from a moving source would arrive at different times & at different amplitudes at each of your ears; BinauralEarlyReflections, like BinauralPan, uses both interaural intensity and interaural time differences but does so in combination with a model of the reflections of sound off the walls of a rectangular room, taking into account both the listener position and the sound source position; SphericalPanner uses filters based on the transfer function of someone's ear for that particular position (the HRTF) and sound sources to the left, right, above, and below a (fixed position) listener and is the only one of the three prototypes that includes elevation (up/down) as well as azimuth (left/right).
Ambisonics is similar to the MidSideEncoder (another prototype you might find useful) which models a pair of microphones — an omni and a cardiod pointing forward. Ambisonics is a model of two or more microphones with more complex polar response patterns. The channels of a mid-side or ambisonic encoded signal can be recombined in new ways to simulate microphone positions that were not in the original recording. Or, if you simulate microphones at the positions of your playback speakers, in theory you can recreate the original signal. Ambisonic enconding/decoding is on our list of features to add to a future release.