... 5,000 Indians pretending to be robots.
5,000 Indians controlling 100,000 robots, though, would represent a huge cost savings in this case.
First, you wouldn't have to hire in the local market, so the labor would be cheaper. The median salary in India is about $350 per month. California's minimum wage (assuming four 40-hour workweeks) is $2,640 per month. So that's almost an order-of-magnitude reduction in cost by itself.
Second, at least half of a delivery person's time is spent in a vehicle going from place to place, and in low-density areas, that increases to maybe more like 80%. Assuming the vehicles are autonomous, the workers controlling the bots won't have to do anything with a bot while it is in transit (and presumably the autonomous vehicles themselves will require only occasional human intervention), so one person will be able to control a large number of bots, taking control of a bot only when one arrives at a destination.
Third, if the bot is autonomous except in exceptional circumstances, the percentage of time that a human will be in charge will be reduced even further.
So if you assume that the cost per hour is lower by a factor of 7.5, that 75% of time is spent driving (no interaction), and that 90% of deliveries are done fully autonomously without human intervention, and that intervention takes only a quarter of the time that it would take to deliver a package entirely by hand, that would mean that a remote delivery driver would cost 0.75% as much per delivery, and that would only improve over time as the reliability of automation improves and the number of interventions decreases.
The end of for-pay non-artistic manual labor is near.