Trips Best Practices

Trip Behavior Fundamentals

When kilo is recording a trip, it collects sensor data to build a map of the environment at record time. When that trip is played back, kilo compares the current sensor data to what it saw initially and tries to match them up. Determining where the robot is in the world using this process is called localization. A high correlation between the original and current maps is considered high confidence for kilo’s localization system.

This confidence check that kilo is doing happens every few seconds throughout playback. It is the most strict at the start of a Trip: kilo needs to be sure it’s starting from the right place. If it passes this strict check, it will start to navigate the path. This is the second most sensitive portion of the Trip for the confidence checking. The longer the robot is on the Trip, the more confident it is that it is on the right track.

Confidence depends on the amount of data points it is receiving from the sensors. Navigating a wide open parking lot is much harder than the hallways of an office. The less objects it can match, the lower the confidence level. If the expected environment does not match up closely enough with the current environment it is attempting to navigate, the robot will exit the trip due to the low confidence. You don’t want a robot wandering around your facility lost.

The image on the left provides a much more clear, permanent landmark to the map-maker than the two on the right.

Improving Trips Success

In general, the closer the environment that kilo is navigating to what it looked like during record time, the better. Endpoints are the most strict localization check, so getting those right can make a huge difference.

General tips

• Have as many large permanent structures or landmarks in kilos field of view during recording as possible.
  • Why: the more nearby objects kilo can match, the higher the confidence it is where it needs to be, and the more likely it is to successfully navigate the Trip.
  • Examples: walls, pillars, shelving, desks
• Minimize movable objects that are likely to move in kilos vicinity during recording.
  • Why: It is easier for the robot to navigate around a new obstacle than to understand that an object it expected to be somewhere has disappeared or moved.
  • Examples to avoid: people, chairs, items being staged for shipping

Recording Good Endpoints

• Start and end trips with enough space that kilo can safely spin in a full circle to capture extra sensor data.
• The best endpoints are in areas close to permanent structures that minimize movable objects.
• If you must start or end a trip in an area without a lot of permanent endpoints, consider adding semi-permanent sensor-height “walls” to the area to give kilo something solid that it can localize off of.