Click here for all current publicly-available radar astrometry, converted to the existing 80-column MPC astrometry format.
MPC does provide some radar data for some objects, but it's not very complete and not usually updated. However, the JPL version of the radar data tends to be complete and reasonably up to date. But I know of no software that actually reads the JPL format. So I wrote a converter program that turns the JPL data into the 80-column format and used it to create the above file. I have a job running on this server which regularly downloads the JPL file and re-converts it.
Output to ADES in either PSV or XML or both would probably be a good idea. (Though in this instance, the punched-card format actually contains almost everything we need, including uncertainties in the data... the most significant missing data are comments and the names of those involved in observing and measuring the data.)
You would not normally compute an asteroid orbit using only radar (you'd take advantage of both optical and radio data), but just for yuks... the following are links to pseudo-MPECs with orbits based solely on radar. The residuals are shown in sigmas.
In some cases, there was a noticeable Yarkovsky effect, which I model with an A1 and A2 parameter (should really only be A2). I just did a few objects for which we have plenty of observations.
You'll notice that in most cases, the residuals are much less than one sigma. (If data are normally distributed, about 2/3 of the observations should be within +/- 1 sigma, 95% within two sigmas, and 99.7% within three sigmas. This is a quick way to tell if data are, in fact, normally distributed.) I've asked about this. Seems the radar guys prefer to overstate the uncertainties in their data.
Truthfully, I'd prefer that uncertainties reflect reality, i.e., "This observation is within one sigma with probability 2/3, etc." The current situation means that radar data are underweighted compared to optical data. I've been computing orbits that use Gaia DR2 data, which has the opposite problem (there are lots of four- and five-sigma errors in Gaia solar system astrometry). Orbits computed using Gaia plus radar have beautiful accuracy, but the totally different concept of how to report uncertainties is annoying.
Theoretically speaking, I could probably analyze the corpus of radar observations and determine a multiplier that should be applied (looks to be about 3) to convert "official radar uncertainties" to "realistic radar uncertainties". And similarly for Gaia-DR2, probably getting a multiplier there of about 0.5.