Skip to content

Latest commit

 

History

History
115 lines (107 loc) · 5.72 KB

FUNTIONALITY.md

File metadata and controls

115 lines (107 loc) · 5.72 KB

Cure funtionality

In this section, we introduce Cure with a walkthrough example. Initially we will get to familiarize with the directory structure used in this repository.

Structure

cure
  |-- ananke (contains third party code for causal effect estimation)
  |-- config (contains search space for Husky and Turtlebot 3)
  |-- data
  |   |-- obs_data (contains observational data used to train the causal models)
  |   |-- bug (contains outlier samples)
  |   |-- results (contains results for each method)
  |   |   |-- cure_results
  |   |   |   |-- platform  
  |   |   |   |   |-- model
  |   |   |   |   |   |-- causal models
  |   |   |   |   |   |-- optimization_snapshot 
  |   |   |-- mobo_results
  |   |   |   |-- platform  
  |   |   |   |   |-- model
  |   |   |   |   |   |-- optimization_snapshot 
  |   |   |-- mobo_results
  |   |   |   |-- platform  
  |   |   |   |   |-- model
  |   |   |   |   |   |-- optimization_snapshot                      
  |-- fig (contains plots used in the paper)
  |-- model (contains pre-trained causal models)  
  |-- src (contains necessary code to implement cure and other baselines)
  |   |-- config (conatins configs used in optimization)
  |   |-- Reval (API used to perform measurements from robots)
  |-- tests (contains necessary code to run and test unicorn)

Modes

  • Simulation mode CURE is compatible with any device utilizing Husky and Turtlebot 3 in Gazebo environment.
    • Debugging
    • Multi-objective optimzation
  • Reality mode the performance measurements are directly taken from the physical robot.
    • Debugging
    • Multi-objective optimzation
  • Debugging mode users can query the root cause of a certain functional and non-functional faults.

Functionality testing

Once the pre-requisites are installed clone the repo and navigate to the root directory using the following:

git clone https://github.com/softsys4ai/cure.git
cd cure

Consider an user deoployed a mobile robot to a new environment and definied the following task specification:

  • navigate to target locations
  • reduced energy consumption
  • higher positional accuracy
  • ensure safety

After deployment. the user encounters energy and task completion faults as the following:

| controller_frequency    |        5.73 |
| planner_patience        |        6.37 |
| controller_patience     |        6.23 |
| conservative_reset_dist |        2.86 |
| acc_lim_theta           |        2.66 |
| acc_lim_trans           |        0.21 |
| acc_lim_x               |        2.88 |
| forward_point_distance  |        0.63 |
| goal_distance_bias      |        5.28 |
| max_scaling_factor      |        0.45 |
| max_vel_theta           |        1.29 |
| max_vel_trans           |        0.58 |
| max_vel_x               |        0.49 |
| min_vel_theta           |        2.69 |
| min_vel_trans           |        0.17 |
| min_vel_x               |           0 |
| occdist_scale           |        0.17 |
| oscillation_reset_angle |        0.23 |
| path_distance_bias      |       33.48 |
| scaling_speed           |        0.35 |
| sim_granularity         |        0.03 |
| sim_time                |        1.65 |
| stop_time_buffer        |        1.13 |
| theta_stopped_vel       |        0.06 |
| trans_stopped_vel       |        0.08 |
| vth_samples             |          20 |
| vx_samples              |           3 |
| vy_samples              |           6 |
| publish_frequency       |        3.81 |
| resolution              |        0.09 |
| transform_tolerance     |        1.15 |
| update_frequency        |         4.1 |
| cost_scaling_factor     |        10.3 |
| inflation_radius        |        1.19 |
| Planner_failed          |         117 |
| Recovery_executed       |          13 |
| Positional_error        |        2.13 |
| Traveled_distance       |      204.86 |
| Mission_time            |     1749.19 |
| Obstacle_distance       |         0.4 |
| Energy                  |      364.04 |
| Task_success_rate       |           0 |

The reported energy=364.04 Wh, Task_success_rate=0, and Positional_error=2.14 meter. To ensure optimial performance of the robot, the user utilizes Cure to determine the root cause of such performance and expects to achieve energy<=40 Wh, Task_success_rate>=0.8, and Positional_error<=0.18 meter. Optionally, the user can also use constraint on Obstacle_distance<=0.25 meter to ensure safety.

To avoid constant human supervision and effort required to perform execution in physical robots, we collect observation data in simulation to learn the causal model using Reval. We already included the observational data in cure/data directory.

Training causal model

The causal model was trained on 1000 observational data obtained using Reval. The following commands can be used for training and saving the causal model. The saved model can later be utilized for both inference purposes and transferring knowledge. We have already included the saved models both for Husky and Turtlebot 3 in the cure/model directory.

python run_cure_MOO.py --robot Husky_sim --train_data data/obs_data/husky_1000.csv

Identifying root causes

Cure can perform debugging tasks such as identifying the root cause of a functional and non-functional fault. The following commands can be used to determine the root causes from the outlier data using the saved causal model. In this example, we have used Task success rate as a functional property, and Energy, Positional_error as non-functional properties. We display the root causes in the terminal.

python run_cure_MOO.py --robot Husky_sim -l --model model/care_Husky_sim.model --outlier_data data/bug/husky_outlier.csv -root_cause  --f Task_success_rate --nf Energy Positional_error