- no true autonomy in chaotic/dynamic environments
- powering these machines for longer timespans is currently impossible (batteries endurance is measured in minutes, not hours)
- high skill level required for operating these machines (again - they're not autonomous in the slightest)
- paradox of complexity: it's easier to genetically engineer tomatoes that can withstand falling to the ground and being swept into a box than it is to engineer a machine that can identify and carefully pick ripe tomatoes as fast, as reliably and as cheap as a human worker
- motion planning and navigating unmapped environments is largely unsolved
- specialised machines are cheaper, more reliable and readily available: it's often more beneficial to modify the process than to program a generalised machine to perform human tasks, e.g. washing machines and dish washers don't need actual hands and tumble dryers don't need to use clothes lines yet all three work fine/better than the human version of the process
- AI safety is still in its infancy (even in the context of "just" motion planning and navigation); there's no such thing as tried and tested safety protocols for fully autonomous robots
- the state of the software (in terms of autonomy) is way behind the state of the hardware
- even Spot is for the most part still just a super low-volume solution looking for a problem
There's a whole ton of unsolved problems:
- no true autonomy in chaotic/dynamic environments
- powering these machines for longer timespans is currently impossible (batteries endurance is measured in minutes, not hours)
- high skill level required for operating these machines (again - they're not autonomous in the slightest)
- paradox of complexity: it's easier to genetically engineer tomatoes that can withstand falling to the ground and being swept into a box than it is to engineer a machine that can identify and carefully pick ripe tomatoes as fast, as reliably and as cheap as a human worker
- motion planning and navigating unmapped environments is largely unsolved
- specialised machines are cheaper, more reliable and readily available: it's often more beneficial to modify the process than to program a generalised machine to perform human tasks, e.g. washing machines and dish washers don't need actual hands and tumble dryers don't need to use clothes lines yet all three work fine/better than the human version of the process
- AI safety is still in its infancy (even in the context of "just" motion planning and navigation); there's no such thing as tried and tested safety protocols for fully autonomous robots
- the state of the software (in terms of autonomy) is way behind the state of the hardware
- even Spot is for the most part still just a super low-volume solution looking for a problem