• Jay Liu
Road Bot >
Design Lessons Learned
  1. Introduction
  2. Planning Robot Tasks
  3. Visualising the Data Coming Back
  4. That's It, for Now

Design Lessons Learned

Associated project: Road Bot

Draft

A growing list of lessons learned from designing interfaces for planning automated processes and map data analysis

July 02, 2021

Introduction


This article builds from the example given in the project page, where road repair crews use robots to identify the areas most in need of repair. Check out that page for user descriptions, workflows, and example data.

Planning Robot Tasks

Lessons Learned in Visualizing the Invisible

Show Robot Intent

In the planning stage, giving the planner a way to see and review what will happen will refine the plan and inspire trust and confidence. Showing system intent is like giving the planner a peek into the future.

Do let the planner see what will happen so they can make necessary changes. In this example, we show the projected search path of the robot.
Do let the planner see what will happen so they can make necessary changes. In this example, we show the projected search path of the robot.

Visualize Robot Constraints

Occasionally, there are hard limits to operational possibilities, such as how far a battery-powered, street traversing robot can travel. Show this upfront to let the planner get it right the first time.

Help the planner get it right the first time without unnecessary trial and error. In this example, we visualize the range the robot can travel before running out of battery.
Help the planner get it right the first time without unnecessary trial and error. In this example, we visualize the range the robot can travel before running out of battery.

Visualising the Data Coming Back

Lessons Learned

Accuracy in perception is hard

Taking the more challenging path of perceptual honesty not only takes more time but can easily be overlooked.

When using GPS coordinates to visualize robot location, for example, using the coordinate “straight out of the box” falsely suggests a location at maximum precision and accuracy. To correctly manage perception, build in a way to visualize error along with the data.

Be honest. Data thrown onto a screen can be a lie without caveats. Seeing this point suggests higher accuracy than what the sensors actually report.
Be honest. Data thrown onto a screen can be a lie without caveats. Seeing this point suggests higher accuracy than what the sensors actually report.

Point the way without getting in the way

The product team may have spent many iterations refining the pipeline to turn raw data into a data product for the data analyst. For example, it could be an algorithm that ingests all the robot data and produces regions with the most serious potholes in a city.

In those circumstances, it’s natural to lose sight of how the user will consume the data product. But unfortunately, putting too much emphasis on the data product can obscure the actual answers the analyst seeks!

In other words, although our focus is creating those polygons, we should not let that detract from the analyst’s focus.

Draw attention to the answer, but don't obscure it.
Draw attention to the answer, but don't obscure it.

That’s It, for Now


I’ll update this page as more thoughts continue to crystallize, but that’s it for now!