I was at Ron's conference talk, and I remember it differently. The spacecraft had a problem, but was not allowed to run on lisp. Hence debugging and fixing it in a repl was impossible.
But Ron wrote a simulator in lisp, and could verify the necessary procedure in his lisp simu. The C code was patched and the spacecraft survived the reboot.
But after re-reading the paper I could have mixed it up, and lisp was indeed in space, and was only killed later.
I often remember his story, and often implement similar simus for my firmwares. And very often such a simu is the best tool to find bugs or to verify patches.
For the reference: 100 million miles is slightly more than 1 AU. It means that one-way communication latency is almost 10 minutes, so the round-trip latency is about 20 minutes.
Of course similar stories, not involving Lisp, abound. So, this story does not, in the end, reveal any unique merit of Lisp, however fun it is as a story.
How many of those stories happened before the DS1, though? It's easy in hindsight to say that our new, non-Lisp system better have remote debugging at least as good as the DS1's. That suggests two (possible) unique merits of Lisp: a repl with the full language environment at one's disposal, as an out-of-the-box feature; and, well, being a trailblazer. :)
Local simulations, testing, etc. A REPL is much easier to connect than gdbstub when the link has 20 minute RTT.
Remote Agent was first for a) being able to be interactively debugged at remote b) actually exercising that, also c) IIRC it was already designed as basis to provide self-healing kind of ability to the system (or I might be confusing it with another lisp system from Ron Garret while he was at JPL)
