Skip to content

Latest commit

 

History

History
87 lines (66 loc) · 3.07 KB

active_learning_workflow.md

File metadata and controls

87 lines (66 loc) · 3.07 KB
Raw

Active Learning Experiment Workflow

By Izzy Brand, Sebastian Castro, Caris Moses, Michael Noseworthy

Overview

Machines:

  • Panda + control box -- the robot
  • RTK -- runs Panda agent, robot interface, and vision
  • Desktop machine -- runs planning server and active learning

RTK Setup

Connecting to the robot

Starting the Franka interface

cd ~/catkin_ws
./franka.sh master
roslaunch franka_interface interface.launch

Interactive Python console to reset the robot position/error state

roscd franka_interface/scripts
python3 -i interactive.py
>> arm.move_to_neutral()
>> arm.resetErrors()

Starting vision server

taskset --cpu-list 10,11 roslaunch panda_vision vision.launch

Starting the Panda agent server

taskset --cpu-list 4,5,8,9 rosrun stacking_ros panda_agent_server.py --num-blocks 10 --real --use-vision --use-planning-server

Desktop Setup

Starting the task and motion planning server

python3 stacking_ros/scripts/planning_server.py --blocks-file learning/domains/towers/final_block_set_10.pkl --use-vision --alternate-orientations --num-blocks 10 --max-tries 2

Active learning (starting from scratch)

python3 -m learning.experiments.active_train_towers --exec-mode real --use-panda-server --block-set-fname learning/domains/towers/final_block_set_10.pkl --n-epochs 20 --n-acquire 10 --sampler sequential --exp-name robot-seq-init-sim --n-samples 100000

Active learning (restarting from existing results)

python3 -m learning.experiments.restart_active_train_towers --exp-path learning/experiments/logs/exp-20210218-161131

To evaluate current progress, run this with the appropriate arguments (tx).

python3 -m learning.evaluate.plot_model_accuracy --exp-paths learning/experiments/logs/robot-seq-init-sim-20210219-131924 --max-acquisitions <tx> --plot-step 1 --test-set-fname learning/evaluate/test_datasets/eval_blocks_test_dataset.pkl --output-fname test

If a tower is mislabeled during acquisition step <tx>, run the following AFTER all towers for step <tx> have been labeled, and acquired_<tx>.pkl has been generated. <tn> corresponds to the tower number integer in the towers file name (0 indexed). <l> is the correct label. If any files from step <tx>+1 have been generated then delete them. (They will be regenerated/trained on restart)

python3 -m learning.experiments.fix_tower_label --exp-path <path> --acquisition-step <tx>
--tower-number <tn> --label <l>

Video collection laptop setup

FFMpeg magic command

Evaluation Workflow

  1. Start franka interface on RTK machine
  2. Start vision on RTK machine
  3. Start planning server on GPU machine (use the learning/domains/towers/eval_block_set_9.pkl)
  4. Run the following to start evaluating where model-type in [learned, noisy-model, simple-model]
python3 -m learning.experiments.run_towers_evaluation --real --use-vision --blocks-file learning/domains/towers/eval_block_set_9.pkl --towers-file learning/experiments/logs/robot-seq-init-sim-20210219-131924/evaluation_towers/cumulative-overhang/<model-type>/towers_40.pkl