Last-mile delivery services are set to become big business, with major investment going into autonomous driving and flying vehicles designed to deliver your takeout or latest Amazon order.

But, while navigating their way along the sidewalks is fairly simple - a number of companies have already experimented successfully with these robots in public - getting right to your door, without relying on pre-existing mapping data and street level imagery, is another issue entirely.

Read More:

• Autonomous robots delivery snacks to hungry college students
• Starship's autonomous delivery robots have complete over 50,000 drop-offs
• Food delivery by drone: Are we there yet?

Six-wheeled robots built by Starship Technologies have already been used to deliver snacks to students on campus at Northern Arizona University, and grocery to beta users in the English town of Milton Keynes. In all, the robots have completed over 50,000 deliveries.

But, as impressive as this is, reaching your doorstep is the next goal - because, come on, who wants to put their shoes on and venture out to the sidewalk to collect a late-night pizza? Exactly.

Along with support from Ford, engineers at the Massachusetts Institute of Technology (MIT) are working to solve this problem by teaching robotic delivery vehicles what to expect when they venture off the sidewalk. For example, they are taught what will likely come next when they encounter certain objects, so upon seeing a driveway they will know this likely leads to a path, which in turn will probably take them to the front door.

Robotic delivery vehicles have to take this educated guessing game-approach, as mapping the space between the street and the front door of every single home is difficult to scale up, and could present privacy issues.

This technique means the robot doesn't rely on maps, and reduces the time it spends exploring the property trying to find the front door by trial and error.

Michael Everett, a graduate student in MIT's Department of Mechanical Engineering, told MIT News: "We wouldn't want to have to make a map of every building that we'd need to visit. With this technique, we hope to drop a robot at the end of any driveway and have it find a door."

As well as understanding how one object - the driveway, for example - may lead to the next, like a path to the door, the robots are being taught to understand natural language. So instead of driving to a coordinates, the robot can see its delivery instructions and understand that the parcel needs dropping at the front door, or outside the garage if that's what the buyer specified. Everett added: "Now we have an ability to give robots a sense of what things are, in real-time."