Test Plan

Fall 2021 (Up-to-date, Oct 4th)

Our test plan until FVD can be found here and 1-pager for FVD here


1-pager for SVD can be found here

Spring 2021 (Updated)

Fall 2021 (Updated)


Spring 2021

Virtual System Validation:

A. Demo Conditions:
  • Location: In simulation
  • Equipment: A computer system or cloud server with Gazebo, ROS and COLA predictive avoidance packages installed
  • Operating area: Simulated 92*64 m factory environment (virtual)

Figure 1: Floor plan for the virtual factory environment

B. Procedure:
  1. The user starts ROS core and simulation
  2. Simulator loads the floor plan for the factory, and launches the visualization tool.
  3. Simulator launches pedestrians and forklifts, following their fixed paths and routines
  4. Simulator launches one robot, and publish waypoints and the static map to robot
  5. Robots plan global paths and start moving
  6. Robots continuously receive localization and noisy obstacle observation from simulator
  7. Robots avoid obstacles according to its predicted trajectories of obstacles
  8. System stops after a certain duration of time

C. Objective & Requirements to Demo:
  1. To demonstrate the simulation system and simulated environment with obstacles
  2. To demonstrate the robot’s ability to plan paths and follow them
  3. To demonstrate the predictive avoidance algorithm (showing robots dodging obstacles using predicted trajectories)
  4. To validate M.P.1-3: Robot speed will be above 0.5m/s and below 1.8m/s if no obstacle is detected
  5. To validate M.P.5: Robots should receive localization and observation in at least 10 Hz, and send back control signals to the simulator in at least 10 Hz

Fall 2021

Real-Life System Validation:

A. Demo Conditions:
  • Location: Any location with flat ground, sufficient lighting and high enough ceiling (over 2.5 m).
  • Equipment: TurtleBot, obstacle action figures/toys, overhead camera, lights
  • Operating area: 2*2 m flat ground

B. Procedure:
  1. Setup overhead camera and place markers on top of robot/obstacles for ground truth gathering
  2. Place robot and one or more obstacles in testing ground
  3. Manually drive robot and/or obstacles
  4. Obtain robot estimation of obstacle positions and classifications for demo duration
  5. Obtain ground truth position data of robot and all obstacles from overhead camera readings for demo duration
  6. Use measurement and ground truth data for visualization/plotting
  7. Evaluate detection (position) and classification accuracy

C. Objective & Requirements to Demo:
  1. To Validate M.P.6: Classify obstacles of interest with mAP of at least 60%
  2. To Validate M.P.7: Detect positions of obstacles of interest within 0.1 m accuracy
  3. To Validate M.P.8: Detect obstacles of interest within a range of 3 m
  4. To Validate: M.P.9 Output results of positioning and classification within 100 ms per frame

Virtual System Validation:

A. Demo Conditions:
  • Location: In simulation
  • Equipment: A computer system or cloud server with Gazebo, ROS and COLA predictive avoidance packages installed
  • Operating area: Simulated 92*64 m factory environment (virtual)

B. Procedure:
  1. The user starts ROS core and simulation
  2. Simulator loads the floor plan for the factory, and launches the visualization tool.
  3. Simulator launches pedestrians and forklifts, following their fixed paths and routines
  4. Simulator launches robot fleet, and publish waypoints and the static map to robots
  5. Robots plan global paths and start moving
  6. Robots continuously receive localization and noisy obstacle observation from simulator
  7. Robots avoid obstacles using all three avoidance algorithms
  8. System stops after a certain duration of time and calculates resulting productivity
  9. System repeats the previous steps in naive avoidance mode without any classification and prediction, and computes nominal productivity

C. Objective & Requirements to Demo:
  1. To demonstrate all three avoidance algorithms (predictive, conservative, reciprocal)
  2. To validate M.P.4: Robot fleet should have productivity increased by >5% when using classification-based predictive/reciprocal avoidance compared to nominal productivity using only naive avoidance
  3. To validate M.P.5: Robots should receive localization and observation in at least 10 Hz, and send back control signals to the simulator in at least 10 Hz