All notable changes to this project will be documented in this file. See Conventional Commits for commit guidelines.
4.33.0 (2023-05-17)
- commands: add
--force-from-scratch
cli arg when running their evaluation to force delete any previous mission information (ae2793c) - metrics: delete evaluation metrics checkpoint at the end of a run to prevent accidentally resuming (6e77b7a)
- metrics: if the action outputs dir is empty and we are not resuming, delete the checkpoint file (b88c03d)
- metrics: support creating and restoring checkpoints of evaluation metrics if resuming runs (bef8c1f)
- wandb: automatically determine whether we should start a new run instead of using an environment variable (82d6ace)
- wandb: include the offline inference version used in the run config (6505c7f)
- wandb: track session ids and success per mission across runs (1260998)
- just manually point to the service registry path (612315b)
- metrics: log the step number instead of trying to modify the config (871ca3f)
- prevent type coercion for the CDFs (within
MissionTrajectory
) (90b9613) - scripts: symlink commands in the prepare-user-data script (e5af1e5)
- settings: boolean whether we resume the run or not (99d3dbc)
- wandb: allow config value changes during a run (bb9499c)
- wandb: column name for the per mission group success rate (e732ee8)
- wandb: saving output files during the evaluation run (ade01ff)
- wandb: set the default success rate to 0 if there are nan's (eec3a00)
- wandb: step needs to be an int (b65310a)
4.32.0 (2023-05-15)
- create flag to enforce/ignore successful preparation steps (4af06aa)
- script: create script that prepares user area for running the offline inference (951080a)
- guard against
InterruptedByNewCommandBatch
(6b8c239) - scripts: automatically setup necessary symlinks (b45cff7)
4.31.1 (2023-05-12)
4.31.0 (2023-04-28)
4.30.0 (2023-04-26)
- T3: add mission 1 (455d023)
4.29.9 (2023-04-26)
- operate microwave challenges (59b8b44)
4.29.8 (2023-04-25)
- coffee unmaker: improve plans, add pot to beans, and iterate layouts (30a6a0d)
4.29.7 (2023-04-25)
- ensure the microwave is empty (a4c454a)
- remove the boss coffee mug from the colour changer missions (eee16b4)
4.29.6 (2023-04-25)
- disabling color variants with microwave challenges (0d8ad3a)
4.29.5 (2023-04-25)
- prep condition for pickup objects from containers (ab9a60c)
- prep goal condition for place object in container challenge (40769c4)
4.29.4 (2023-04-25)
- disable start colour variants for the colour changer challenges (a1e39cb)
- remove objects that cannot be color-changed (66451f1)
- update plans for the challenges (b4269d1)
4.29.3 (2023-04-25)
4.29.2 (2023-04-23)
- target object was not flagged as ambiguous in key (5579cda)
4.29.1 (2023-04-23)
- high level keys and plans (b1dfde9)
- include from_receptacle in breaking challenges (0f236ba)
- remove the from receptacle from high level key in break challenges (8db7588)
4.29.0 (2023-04-23)
- desks should not be unique (9f3f1f2)
- do not use the colour changer to make the object the same color (80d9a90)
4.28.0 (2023-04-23)
- check that the gravity pad contains the plate during prep subgoal (6ec8b31)
- disable color variants for fridge/freezer challenges (1778d72)
- update room/colour was failing because there was no 0th index (a57bb0c)
4.27.0 (2023-04-23)
4.26.0 (2023-04-23)
4.25.0 (2023-04-23)
- add breakroom table to required objects list (4e6d8b6)
- dont make the spawned object into a required one (987a287)
- replace the spawned object with the printer cartridge (172fb7a)
4.24.1 (2023-04-22)
- expliticly ensure at least the first subgoal has been completed (f61f810)
- return after logging the failed trajectory (940d400)
4.24.0 (2023-04-22)
4.23.0 (2023-04-22)
- add goal for prep steps (f3db7ec)
4.22.0 (2023-04-22)
- add goal for prep steps (7742128)
- add goals to ensure the "objects in containers" missions are setup correctly (606a4c3)
- mark the run as failed if the preparation steps did not succeed (4830023)
- mark the run as failed if the subgoal success is 0 (f3659b3)
- disable color variants usage for pickup and place missions separately (b941892)
4.21.0 (2023-04-22)
- add the
target-object-is-ambiguous
high-level key (f534502)
- update high-level key config within wandb tracker (3bbc069)
4.20.0 (2023-04-21)
4.19.1 (2023-04-21)
- wrong receptacle for target object (c45a898)
4.19.0 (2023-04-21)
- make it easier to generate trajectories in a different folder (4215037)
- make the output dir for generated trajectories (fcedc65)
4.18.0 (2023-04-21)
- enable carrot machine challenges (321d7a6)
- kill the unity instance before trying to restart it (1274ae2)
4.17.0 (2023-04-21)
4.16.1 (2023-04-21)
- always remove objects from microwave (ea0dbcd)
4.16.0 (2023-04-21)
- include stacked objects in the high level key (3f937e7)
- remove the
change_color
instruction action (d53cbaa)
4.15.1 (2023-04-21)
- color changer preparation step (cc548ec)
4.15.0 (2023-04-21)
- use color variants for both boss mug and normal mug (a7b5371)
4.14.2 (2023-04-21)
- incorrect function call (af355fd)
4.14.1 (2023-04-20)
- attempt to handle the 408 connection error (467f5c7)
4.14.0 (2023-04-20)
- include unity logs with each run (bdea65e)
4.13.3 (2023-04-20)
4.13.2 (2023-04-20)
- goal conditions for the operate printer challenges (151647c)
4.13.1 (2023-04-19)
- disable the carrot machine challenges (4badcef)
4.13.0 (2023-04-19)
- add more helper methods to structures (893f0b6)
- change starting room to robotics lab (518425a)
- let object instance ID's end in a * (f3d7713)
- make the argument name be more descriptive to what is needed (93e1760)
- operate printer challenges (82ce622)
- operate printer: make sure robotic arm is out the way (1dca91b)
- operate time machine with carrots (1f85703)
- printer cartridge name in the preparation plan (b8250fa)
- typo (0355636)
- validator condition for object instance id suffix (d8fb10f)
4.12.1 (2023-04-19)
4.12.0 (2023-04-19)
4.11.0 (2023-04-19)
4.10.0 (2023-04-19)
4.9.0 (2023-04-19)
4.8.0 (2023-04-19)
4.7.0 (2023-04-19)
- support setting wandb group name from run command (9b72cc8)
4.6.0 (2023-04-18)
- Add more pickable objects for fridge/freezer (#18) (baf6850)
- save the mission trajectory file and the output file to wandb (211bd8c)
4.5.0 (2023-04-18)
- log the experience hub version with the wandb run (45c17fb)
4.4.1 (2023-04-17)
- command that starts the experience hub (f187fef)
- command used to kill the experience hub (5e8bd3d)
- if failed to go to a random viewpoint, just go to the first one in the room (bb3d652)
- use 2 workers to hopefully stop the experience hub from crashing (daa17a0)
4.4.0 (2023-04-17)
- also send subgoal completion success rate to wandb per session (3a789f0)
- set the session id as the run name (1960040)
4.3.0 (2023-04-17)
- use subprocess to run the experience hub because its easier to consistently kill (bfaeea2)
- disable fast mode (d894ef2)
- disable look actions in random walk (6060c33)
- increase number of healthcheck attempts for experience hub (a6737a2)
- plan for placing objects in container (f522c26)
- remove initial contained items from various containers (d53b7a0)
4.2.0 (2023-04-16)
- track progress of generated trajectories on wandb (cb5c1ab)
- set experience hub timeout to be stupidly high (9302cca)
4.1.0 (2023-04-16)
- store any remaining utterances that have not been sent to the action outputs (2bf68b7)
- break out the loop if all the goals are complete (5b8a18f)
- break out the loop if the goals have been completed (5abe3b1)
4.0.1 (2023-04-16)
- make sure the breakroom table exists, and has space for the preparation items (34a10cd)
- preparation plan for fill object in sink challenge (86a96b0)
4.0.0 (2023-04-16)
- migrate challenges to using state conditions
- "and" is no longer allowed (1ca4655)
- ensure the reimported CDF is same as the original CDF (df7a6b8)
- go to a random viewpoint that actually exists in the current scene (f4a21cb)
- high level key action for "fill object in sink" (af4829b)
- remove the duplicated running of preparation steps (4db9b7e)
- use join and close when killing the experience hub process (3c18b58)
3.2.0 (2023-04-14)
- do not send utterances to the arena if all the goals are complete (b8cc13c)
- make it easier to kill the experience hub when something goes wrong (7e375fe)
- fill the object challenge goals (8543bb2)
- include a preparation plan to be able to pickup objects (#16) (d86301a)
3.1.2 (2023-04-13)
- remove the trajectory batching (a467340)
3.1.1 (2023-04-12)
- do not randomise trajectory run order by default (3e16279)
- get rid of the progress bar (133b049)
- remove old "kill arena" command (23e4cda)
3.1.0 (2023-04-12)
- restart the arena after every 10 sessions (0f9fcdd)
- improve the styling of the progress bar (3191a34)
- update the progress bar after sending the utterance (5d698c3)
- update the progress bar more (4ecce70)
3.0.0 (2023-04-12)
- use a new structure for the high-level keys
- do not re-run missions that have already been run (1720664)
- use a new structure for the high-level keys (2116539)
- add more state names to the arena constants (79365b2)
- use kebab-case when converting high-level key to string (2e2d3e8)
2.13.0 (2023-04-12)
- print a table of num challenges per high level key (4364dbd)
2.12.0 (2023-04-12)
- separate the trajectory generation from the trajectory running (4b9cd0d)
2.11.0 (2023-04-12)
2.10.0 (2023-04-12)
- ensure each object-related key is a 'readable name' (in the
HighLevelKey
(0cb022e)
2.9.1 (2023-04-12)
- explicitly forbid unsupported keys from the
HighLevelKey
(aca0079)
2.9.0 (2023-04-11)
2.8.0 (2023-04-11)
- add challenge to clean a plate in sink (392518a)
- object instance ids for the sink and plate (9354196)
- use deepcopy on the sink and create a trajectory for every layout (0b004a5)
2.7.0 (2023-04-11)
- lowercase the session ids (except for the prefix) (eac8202)
- separate the readable name from the object key when building challenges (35fbde2)
- use readable names for the keys (9c30b00)
2.6.1 (2023-04-11)
- add more error types to ignore when randomising start position (f866cdb)
- go back to using
str
forstr
-based structures (8a62425)
2.6.0 (2023-04-11)
- add challenges for broken bowls and coloured bowls (43cb76c)
- improve the progress bar for the challenge validator (9e9e2d2)
- only use the colour changer colors (4775e46)
- shuffle the order trajectories are generated in (33ea2fa)
- support using RNG for the CDF scenes (with
floor_plan
) (b7e1b4d) - validate cdfs from generated missions (062178c)
required_objects
key within theCDFScene
(fe3bcd1)- add
__str__
methods for theObjectId
andObjectInstanceId
(9b764be) - change CDF
floor_plan
validation to allow for"-1"
(d12bfe2) - object ids for broken cords and computer monitors (33aaffa)
- remove the duplicated object state (3d92f22)
- send actions to randomise start position one-by-one (8b31f76)
2.5.0 (2023-04-10)
- generate all the pickup from fridge/freezer missions (e263c93)
2.4.0 (2023-04-10)
- add
insert
andvendingmachine
actions to the high-level key structure (a6486a2) - add command to print the high-levels keys that we have challenges for (051043e)
2.3.0 (2023-04-09)
- randomise start position before challenge begins (f78a9fb)
2.2.0 (2023-04-09)
- add 'pickup apple from fridge' mission, with coloured variants (f431757)
- register challenge to pick up (coloured) apples from an open fridge (43fd6ed)
- support generating challenges from other challenges with minor modifications (1ffbc1e)
- make sure the
isColorChanged
key doesn't exist already to prevent duplicates (9555241) - replace property setters with explicit functions (2d4038f)
- using
*-is-container
when parsing high-level keys from string (6c78d32)
2.1.0 (2023-04-09)
- improve generalisability of challenge builders for less boilerplate (fe24100)
- add tests and make sure the generation process works (1a6f6ac)
2.0.0 (2023-04-08)
- be able to generate missions from high-level keys
- add more missions for other layouts (1c5615d)
- add option to send dummy actions when validating cdfs (0de8f4b)
- be able to generate missions from high-level keys (1772f92)
- include progress to more clearly know the overall progress (e135740)
- optionally add randomness to the session id name (c0c43e2)
- upload trajectory results to s3 (77d1532)
- use cloudpathlib to upload all the metrics to S3 (986b1ce)
- create all parents for the metric output file (41d9372)
- created session id needs to not have slashes (644268b)
- created session id to be in the form
T.DATE/KEY-UUID
(e94d2ff) - env var key to enable the offline evaluation mode (5380a4b)
- generated session id that is valid as a path and uri (c2a033d)
- lint issues (ad3527d)
- set a long timeout for experience hub, which gets overridden by the settings client timeout (30b45c1)
- simplify the provided session id prefix (70c5971)
1.5.0 (2023-04-06)
- add command to validate cdfs within a dir (8b65c55)
- add flag to enable offline evaluation mode in experience hub (4514467)
- add missions for picking up from freezer (8108cf1)
- clone the experience hub into the storage dir (22b91a7)
- create new structures for the challenges and trajectories (1e12c0b)
- improve how generating trajectories are to be run (d3b8979)
- improve the first attempt to make the CDFs (c19be24)
- just use a single high level key for each mission (788ea34)
- kill command for the arena (605863b)
- missions for pickup from fridge (b74601c)
- settings need to exist before being able to run things (951aa38)
- the service registry path (eb8c5c5)
1.4.0 (2023-04-05)
- be able to validate cdfs (faa6722)
- if the arena/experience hub are running, do not try and start it again (cde1077)
- make sure the experience hub dies (5094187)
- run command with the new experience hub version (6ad167b)
1.3.1 (2023-04-04)
- context managers for the controllers/orchestrators (fc27d12)
1.3.0 (2023-04-02)
- simplify run commands and add in the backend for the web tool (4b4ee95)
1.2.0 (2023-04-01)
- add options to run for subset (2391594)
- create separate module/script to run the background services (ec19cd8)
- dump all metrics so they can be stitched together again (d8366ce)
- improve metrics calculating (1bfa793)
- include CDFs (d22fe14)
- just everything from running the eval before report submission (07b000e)
- only evaluate missions that have not been evaluated yet (453ceb9)
- run eval on single gpu (bf16694)
- set instance range in settings and send to s3 when done (b126ffc)
- improve logs and healthchecks and responses (849ba61)
- just make loads of changes to make it actually work properly (041d40c)
- just make loads of changes to make it actually work properly (2a88df5)
- make sure we do all instances and dont miss any (f4901b5)
- use the new arena executable (c8a287d)
1.1.0 (2023-03-19)
- handle lightweight dialogs within the context of the actions (8647fe7)
1.0.1 (2023-01-24)
- only download and prepare T2 validation data (54f716e)
- add arena wrapper from ml toolbox (9837f96)
- add code to orchestrate the test (bc17922)
- add experience hub dir to settings (d292d1b)
- add log points (bac27af)
- add metric logging for the evaluation (c0f8ee7)
- add run command (0830cc5)
- add script to launch xserver (21389ce)
- add scripts to download the mission data (ef568c6)
- add terraform config for creating the instance (4ab0e6c)
- automatically prepare the file system for the evaluation (397018b)
- automatically update permissions and start xserver (37a39fe)
- create the settings file (1b45fcd)
- disable client timeouts on experience hub (ed5551d)
- install multiprocessing logging (b767f43)
- set ssh key to the ec2 key (e4a98bd)
- setup repo (39949b5)
- setup venv and prepare trajectory data in user-data script (4ccf28d)
- stop docker containers on exit too (a369b42)
- use experience hub for access to the storage/docker configs (9249bbe)
- use loguru for logging in arena_orchestrator (c9fc3ba)
- use rich logging (557b006)
- add catch for timeout on healthcheck (0bbab99)
- allow pickle when loading data (7ac9ec0)
- change port for the experience hub to run on (f211885)
- copy the arena deps as ubuntu (91f8c90)
- create session id directory for auxiliary metadata (214ccb6)
- creating the storage dir for cloning the experience hub (8cc6dbf)
- disable multiprocess logging (f8cdc4a)
- do not create a symlink (70cb36f)
- do not force download the models if they exist (5b8949e)
- do not run process as a daemon (7ec2f2c)
- do not start xserver within the user-data (700f440)
- do not try to setup the python env on launch - it wont play nice (9380128)
- explicitly define the args to run the controller api (a97ddb4)
- explicitly disable observability and production (79b43cd)
- formatting (b21c550)
- improve orchestrators start order (defb686)
- lint errors (ada1332)
- method order in class (4a49a24)
- model storage dir (ced8940)
- only need about 10 retries before it should be running (61f6e20)
- order of setting orchestrators up (f4949ef)
- re-able running as a daemon (eceb2f4)
- remove dialog actions from the experience hub response actions (e7f1d50)
- remove the xserver module - its not needed (f1836b4)
- send dummy actions when loading the game (c152998)
- set arena env vars within the run (5edb809)
- set the appconfig to a dataclass so that it hopefully loads the env vars (31624c9)
- set the envvars outside the function (4192e36)
- type error because im pickling so we dont care (5f0c3a3)
- types for paths in case they dont exist already because we create them (05d565d)
- unzip path for cdf data (a035e7c)
- use the httpx client when making the request (5897d15)
- use the settings to hopefully run the thing (db07a39)