Skip to content

Latest commit

 

History

History
135 lines (128 loc) · 5.63 KB

CHANGELOG.rst

File metadata and controls

135 lines (128 loc) · 5.63 KB
Raw

Changelog for package hri_engagement

2.0.1 (2024-08-21)

  • rename diagnostics msg to match documentation (and diagnostic_aggregator) categories
  • Contributors: Séverin Lemaignan

2.0.0 (2024-08-21)

Port to ROS 2

  • fix test dependency
  • workaroud missing dependecy in tf_transformations
  • improve lifecycle shutdown handling
  • linting
  • add license
  • exclude obsolete tests from linters
  • cleanup dependencies
  • port synthetic data tests
  • use sim time in listener; porting fixes
  • refactor parameter variables
  • enable synthetic and linting tests; relocate obsolete tests
  • fix not properly destroyed interfaces on deactivate and test tear down
  • launch: use get_pal_configuration from launch_pal
  • wip unittest
  • WIP on the unit tests
  • upstream pyhri bug fixed; can properly destroy hri_listener
  • test runs without crash, replay bag files, but does not actually test hri_engagement
  • fix some crashers
  • initial scaffolding for ROS2 unit-tests Unit-testing infrastructure mostly borrowed from asr_vosk
  • convert test bag files to ROS2
  • Initial ROS2 port Node runs; no testing yet, as tests need to be ported first While here: - add diagnostics publishing - expose parameters for observation window duration and note rate
  • Contributors: Luka Juricic, Séverin Lemaignan

1.0.1 (2024-01-10)

  • add missing dep on rostest
  • Contributors: Séverin Lemaignan

1.0.0 (2023-12-05)

  • [test] disabled simple 'disengagement' bagfiles The current version of the algorithm consider these people to be partially engaged, thus failing the test
  • [doc] fix/update/complete README
  • tune the algorithm for more plausible results Tested with myself in front of the webcam While here, only publish the 'ENGAGE_WITH' intent when we first become ENGAGED
  • change how distance is used to compute engagement Instead of simply dividing mutual gaze by distance, use a more complex log-based version of the distance that goes down 'gently' towards zero
  • add param to set a maximum distance to even consider engagement By default, 4m
  • {hri_engagement.launch -> monitor.launch}
  • {engagement_node.py -> engagement_monitor}
  • add synthetic tests
  • Add two unittest for hri_engagement - Add a new test folder called complex_bags with bags that do not pass the test - Add a new test folder called simple_bags with bags that need to pass the test - Code refactoring
  • migrate code to pyhri
  • add ROS parameters to configure field of view and social engagement threshold
  • reimplement the core algorithm to use the Visual Social Engagement metric From: Measuring Visual Social Engagement from Proxemics and Gaze (Webb and Lemaignan 2022) While here, adjusted several constant for more reliable behaviour
  • Contributors: Luka Juricic, Séverin Lemaignan, antonioandriella, saracooper

0.1.0 (2022-08-02)

  • Add ReadMe
  • remove engagement callback rate in params
  • [DEBUG] duplicate call to method compute_engagement
  • code refactoring - improving code structure - add sellion_link as robot_gaze_frame - add timer to track tf availability (10 secs) - set engagement to unknown when person not active - remove timer callback for engagement status
  • code refactoring - design new class structure - define an aux Person class that computes the engagement_status and publish it on a topic. - edit EngagementNode class that now only checks for the active persons and for each of them initialises an instance of the class Person. - migrate code from tf to tf2 - add var is_registered to Person - simplify compute_engagement method
  • [feat] engagement node working with >= 1 human the hri_person_manager recognises a human and assign them a unique id. A dedicated callback has been introduced to publish the engagement status at a given frame/rate.
  • [feat] engagement node working with 1 human It publishes the status of an active human as: ['disengaged', 'disengaging', 'engaging', 'engaged'] based on their previous status history
  • Changed reference frame for resting head position
  • Publishing PoseStamped on /look_to_wildcard /look_to_wildcard topic expects PoseStamped object, not PointStamped. Removed the topic latcher since it is no more required
  • Reduced publishing rate on /look_at The node now publishes a PointStamped object on /look_at only if: -- The engaged face has moved more than a specific threshold value from the last position the eyes have moved for -- The state machine goes from state 2 (Retrieving engagement) to state 0 (No engagement). In this case the robot will go back looking in front of it and eyes need to be repositioned.
  • Fixing attention geometry and new debug tools -- Debugged formula in the cone-of-attention section -- It is now possible to visualize the current state machine value as a semaphore in rviz
  • Publishing engagement information for head turning Through topic_tools relay, the information about the point the robot should look at gets published also on the /look_at_wildcard topic, involved in head turning.
  • introduced robot head frame parameter
  • code refactoring -- removed comments and unused dependencies from package.xml -- removed comments and unused dependencies from CMakeLists.txt
  • fixed active users publishing -- added function update_active_users to track all the ids that have been active at least once in the last time_thr sec -- code refactoring
  • code refactoring
  • Modifying the structure representing the close faces
  • publishing ids of the active users
  • Initial commit
  • Contributors: Antonio Andriella, Séverin Lemaignan, Lorenzo Ferrini