The main constraints in small combat bots are weight > space > price (within reason).
Ideally I’d want enough torque for 2-4 wheels to drive a 1lb or 3lb bot with enough overhead to push around the other bot (same weight). This requires winning the traction game (polypropylene on plywood or steel). Easy if the opponent uses foam wheels, or is 2WD vrs my 4WD (since they’re giving up part of their traction to their bot dragging). Complicated by uneven damaged floors, and dust reducing friction.
4mm-5mm shaft for durability (3 is acceptable but a little risky).
Direct drive is ideal, since it reduces space, and allows for a hub motor (even more space savings).
Direct drive RPM topping out around 3-4k (for a 1.5" wheel) at 4S (200-300kV?).
Motor driver that can ramp up and down to full speed reliably regardless of how aggressively the throttle is used (this is my hope for FoC+sensor handling the low end of the throttle). Currently I manually insert some dead time on throttle reverse, to prevent the blheli_s from freaking out, since going from 20% throttle, to -20% doesn’t make its sensorless configuration happy.
I really like the idea of all-in-one motor drivers in the motor mount, especially if there’s an integrated sensor configuration. I’m keeping an eye on this dual hall sensor thread: 40 cent magnetic angle sensing technique since it seems like it might be a good fit.
I think it’s all doable, however it seems likely that the result will be $150 custom machined, hand wound and hand made per motor 
The other thing I’m interested in is the possibility of doing a 1lb walker bot. This is very impractical because currently it would have to use servos, which are too fragile and can’t survive a limb being backdriven by a weapon impact. If there was a small and cheap enough 4-6 gimbal motor configuration that could move a biped, or 6-9 for a tripedal bot it would be really neat. A “true” walker gets a weight bonus of 200% (allowing a 2lb walker in a 1lb match). True being typically defined as “not involving continuous rotation in a walking pattern”, so excluding cam based mechanisms that simply convert continuous rotation into linear motion. Those typically get half the bonus (“shufflers”), similar to vibration bots and most gyro-walkers. The limitation seems to be that any mechanism must involve motors reversing direction as part of the walking motion.
As a third topic, my boards arrived! I
I may attempt soldering the first one today.