I'm all in favour of KISS. The 37 command bytes are more less arbitrarily chosen, so recoding them with a greater hamming distance is the simplest solution I can think of. Sending 2 bytes actually requires more complicated code, although it's obviously still relatively simple.

The goal is to maximise robustness in an industrial environment where ESD and fast transients might cause bit errors. As we don't know the BER, we are just looking for generic ruggedness without added cost.