What we badly need in the military department is to field large forces of remote-controlled, "dumb" disposable ground battle drones ("groundbots")--smallish plastic and metal robots run from far away over a comm link. Using drone groundbots will save the lives of many soldiers, save money, and put lower-tech adversaries in a position of extreme disadvantage.
Since the Predator program has already devastatingly proven the effectiveness, efficiency, economy, and practicality of remote-drone combat, extension of this technology to ground drones is inevitable and highly desirable. While some groundbots are already in use for specialized missions (e.g. mine clearance), they aren't being used in direct combat as proxy soldiers. One problem is the unwise attempts to make some of these bots autonomous--which is to say: expensive, complex, unreliable, and dangerous to our own troops. Far more useful are less complex, dumb drones that require full human control over a comm link--pure "proxy" drones with no onboard decisionmaking intelligence or autonomy.
Specialized hunter-killer drone "squads" designed to attack terrorists in the hills could likely be put in the field with existing technology. Unlike treaded "toy tank" designs currently in use, low-profile crab-like designs with multiple legs/claws will be better able to climb and scramble through chaotic, rocky terrain such as Afghan mountainsides. A high-profile treaded bot that can only operate on roads, clear flat terrain, and gentle slopes is useless in mountain combat.
For example, consider a 5-unit hunter-killer squad of battery-powered crab-like or lizard-like drones, each operated remotely by a dedicated specialist 3 miles away via links to a Predator circling overhead, which in turn relays commands to the units on the ground, and relays close-in video and data from the battle drones back to the operators.
Released at the edge of an area containing enemy positions, the units spread out and scuttle toward the enemy positions, moving one at a time under direct control of their remote operators. Two small basketball-sized units, heavily camo-ed and gillied up, are "dazzle and distraction" bots, with laser dazzlers, high-power flashlamps, and loudspeakers. When their operators observe enemies in the rocks via the bot's cameras, they target them with the dazzle lasers and the bots move toward them emitting loud alarming audio screams and flashes. While the enemy are distracted by these units, a large, obvious, armored "target" unit also advances on their position, firing shotgun rounds. Its role is to draw fire and attention, and to appear to be the main threat. The dazzle bots alternately move, hide, and dazzle. But while this is going on, the operators of two camouflaged "killer" units are able to spot individual enemies, position their bots slowly, aim carefully with accurate rifle weapons mounted on the bots, and take down the enemies. Some "target" and/or "dazzle" units may be lost or damaged in such exchanges, but these will be cheap and disposable by design.
What does "cheap" mean? Well, the cost of maintaining a US ground troop in Afghanistan has been variously reported as between $850,000 and $1.4 million a year. If some of those soldiers can be replaced, or intermittently proxied, by small drone robots costing maybe $20,000 each in volume, there is a cost advantage, especially if soldiers' lives are saved. Of course, there has to be at least one operator, probably several, and probably a Predator relay in the sky, and these things have to be factored in as well along with their support costs. (Why $20,000? Well, that happens to be almost exactly the unit cost of a shipment of bots that were actually sent to Afghanistan in 2011. But also consider that car companies can deliver much larger, more complex vehicles to US consumers at retail prices in that range. No matter how gold-plated the military makes them, how much can it cost to build a small unit with a battery, some electric motors, a comm link, a camera, a gun or laser, and related remote control systems? The great advantage of remote proxy drones is that they require no autonomous intelligence, which is where robots get expensive and problematic.)
Then too, simply locating the enemy and laser-tagging their approximate location would help the operator of a Predator overhead to further examine, and perhaps target the position with a missile, skipping the entire close-combat drone scenario. So the above-described "laser crabs" might be the quickest way to leverage drone technology onto the ground--just build units designed to scuttle over rough terrain, alternately moving and hiding and sending ground camera images to their operators. Once enemy are seen, on goes the laser spot and the air drones take over. (Though, at around $50-70K/missile, this option might turn out to be more expensive over time. The missiles could cost more than the drones.)
Remotely running dozens of small coordinated bot-squads in this way could become the new normal for ground combat in the 21st century. Robot troops, over time, will likely make direct combat between human soldiers less common, because of the economics and the lives saved. But in the meantime, advanced nations fielding thousands of disposable drone battle-bots will put Taliban-like low-tech adversaries at an increasingly huge disadvantage, just as the Predator drones have done. Unfortunately, the more distant future then probably holds the grim prospect of superpowers squaring off with drone armies in the millions, but you can't stop progress.