GPS vs. Odometry: How Roombas Know Where They Are

Google Maps is amazing. It puts a blue dot on your house within 5 meters. But “5 meters” (15 feet) is huge. If a Roomba was off by 15 feet, it would fall down the stairs. If an FTC robot was off by 15 feet, it would drive into the audience. Also, GPS satellites (12,000 miles up) cannot penetrate indoor roofs.

So, how do indoor robots navigate? Dead Reckoning (Odometry).

Dead Reckoning: “I Counted Steps”

Imagine you are blindfolded. You take 10 steps forward. You turn 90 degrees right. You take 5 steps. You have a rough idea of where you are relative to the start. This is Odometry.

Mechanical Odometry (Encoders)

Robots use Encoders on their wheels. An encoder counts the ticks of rotation.

• 1 Tick = 0.01 inches.
• The software calculates: “Left wheel moved 1000 ticks. Right wheel moved 1000 ticks. Therefore, I went straight.”
• “Left wheel moved 1000. Right wheel moved -1000. Therefore, I spun in place.”

The Tragedy of Drift

The problem with Dead Reckoning is that errors accumulate.

• If your wheel slips on a carpet for just 0.1 seconds, the encoder counts ticks, but the robot didn’t move.
• The robot thinks it is at position (10, 10). It is actually at (9, 10).
• After 2 minutes, these tiny errors stack up (“Drift”). The robot thinks it’s in the kitchen, but it’s actually stuck under the couch.

Visual Odometry (The Eye)

Newer robots (like the Roomba j7 or Dyson) use cameras.

• They look at the ceiling corners. “I see the lamp.”
• As they drive, the lamp moves.
• They triangulate their position. This mimics human navigation. We don’t count steps; we look at landmarks. In FTC, we combine both. We use Encoders for fast updates (1000Hz) and Vision (AprilTags) to “reset” our drift every few seconds.