merge in master into current working branch #1737
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
* callplanner * init nsrt * assertion * tab * minor fix * autoformat * fixes * small fix * minor change * more fixes * more fixes * comment --------- Co-authored-by: “matrixbalto” <“[email protected]”>
…learning. (#1672) * Initial commit. * Add geo and vlm approaches to offline data methods considered in _learn_nsrts() for annotations. * Remove pdb statement. * Run yapf. --------- Co-authored-by: Nishanth Kumar <[email protected]>
* Initial commit. * Note true vlm atoms in utils.abstract() to make it easier to debug. * Run autoformatter.
* clean implementation of renaming ft. ashay! * fix checks
…cates (#1682) * Update pour option init * fix yapf --------- Co-authored-by: yichao-liang <yichao-liang>
* see if we can fix mypy issue(s) * yapf * fix mypy * should be gtg
* making progress * planning for new kitchen goal(s) works! * fix mypy * yeet
* Initial commit. * Undo debugging thing. * Adjust goalhack predicate and goal in tasks. * Debugging changes. * Run autoformatter and get some of the checks to pass. * Update BurgerEnv test after the grid was changed from 4x4 to 5x5 and task variation was added. * Run autoformatter. * Update tests to get better coverage. * Add comments and make small edits to pass checks. * Get more checks to pass. * Remove pdb statement. * Try to fix typing issue. * Fix incorrect usage of cast. --------- Co-authored-by: Nishanth Kumar <[email protected]>
* Initial commit. * Add printing of traj number in labeling progress output. * Get checks to pass. * Add to test to get full coverage. * Refactor for cleaner code. * Fix typing issue. * Run autoformatter. * Add comment explaining use of multithreading. * Run autoformatter.
* clean * cleaner * rename open * formatting * cleaning * deleted print --------- Co-authored-by: “matrixbalto” <“[email protected]”> Co-authored-by: Nishanth Kumar <[email protected]>
* doorknobs added * testing * linting * more linting * refactor * more refactoring * small edit * cleaning * refactoring options * more refactoring * linter * minior fixes * coverage * more linting --------- Co-authored-by: “matrixbalto” <“[email protected]”> Co-authored-by: NishanthJKumar <[email protected]>
* making slow progress * making good progress; almost there! * good to go! * feature-complete! * good to go * fix checks * fix tests
* modified gnn option * formatting * type def * all checks passed
* Add compatibility for LLMOpenLoopApproach to use GEMINI --------- Co-authored-by: yichao-liang <yichao-liang>
* draft generate dataset with vision api. use with flat --vlm_predicate_vision_api_generate_ground_atoms True * vision api baseline running * fix pylint * works with multiple demos * yapf" * fix mypy * improved regex * uncomment sample * restore accdentally deleted files and changes to docstrings * fix pylint * add test for lfd-vision-api * add pylint: disable=exec-used * yapf * add prompt file for pytest * fix tests * yapf * pylint style * git add tests * yapf * fix test * refactor * yapf --------- Co-authored-by: yichao-liang <yichao-liang>
* init stuff * making progress * progress! * making progress! * progress! Still need to construct few-shot prompt, but basically there! * imgs are prepped; got to make actual prompt and query model now * ready to go! * everything except tests done! * works for envs with continuous params now! * woops - mypy * yapf * testing utils done * ashay comments
) * Initial commit. * Add first-draft of support for cropped images that focus on relevant objects. * Update approach to pass in full and cropped before/after images during labeling. * Update prompt and add some debug statements. * Add option to compute preconditions via soft intersection in cluter and intersect. * Investigate extra operators from label errors that cause us to fail on test tasks. * Start adding some support for goal variation in BurgerEnv. * Add a GoalHack for the third goal in BurgerEnv and fix set of marks to not extend past the edge of the grid. * Add debugging code via a manually specified plan. * Initial commit. * Update nsrts and options for burger_no_move. * Fix burger_no_move nsrts and options so that basic goals are achieved by oracle. * Get one basic generalization result in burger to work. * Edit goals towards a different generalization result. * Progress. * Achieved next generalization result. * Fix VLM atom proposal to include all groundings of each proposed predicate. * Update get_tasks(). * Get 2/3 categories of generalization to work. * Get all three kinds of tasks to work. * Refactor goalhack predicates and ensure BurgerEnv still works. * Fix some formatting issues. * Fix more linting issues. * Get formatting, linting, type-checking tests to pass. * Get mypy tests to pass. * Run autoformatter. * Start to address comments. * Get some tests to pass. * Add an option to skip hill-climbing and select all debug predicates directly. * Run autoformatter. * Progress towards getting burger_no_move to work after changes. * Update debug VLM atom proposal to use existing code that was there to handle that case. * Fix handling of all groundings of VLM atom proposals to avoid labeling them when their objects are not the trajectory being considered. * Remove comment. * Run autoformatter. * Fix typo in comment. * Replace original atom labeling file. * Remove space. * Add unit tests for grammar search debug predicates and soft intersection in cluster and intersect. * Fix burger text. * Include cropped image into prompt only when flag is set. * Add test for including cropped images in ground atom data generation. * Run autoformatter. * Add more tests for include cropped images. * Add test for ImageOptionTrajectory. * Update tests for BurgerEnv. * Write tests for BurgerNoMoveEnv. * Update burger no move test. * Update burger no move tests. * Finish burger no move tests. * Fix broken tests.
* should be good to go! * revert unnec ipdb * fixes test coverage * fix circular importing issue
* fix train and test generations in burger! * minor fix * Burger env task gen fixes ashay (#1710) * Initial commit. * Update prompt to tell VLM to use previous value for atoms that it doesn't expect to change. * Fix typo that allows combo_burger to run. --------- Co-authored-by: Ashay Athalye <[email protected]> * yapf * revert unnec change * fix burger tests --------- Co-authored-by: Ashay Athalye <[email protected]>
* new baseline should be good to go! * fix bug with renaming predicates * revert unnec newlines * fix linting * testing * fix testing bug * yapf
* fix goal predicate setting for burger * add holding to all goal predicate expressions in burger * testing and linting
* init commit * test local command * change setting name * renaming * adds correct offline data method * add plan length measurement to metrics for FD * adds some metrics + finishes up yaml file for pred invention vlm! * progress * push to run job locally * push again * add flag * update burger goal preds * include givens false - aha!!! * fix * fix 2 * updates to enable running baselines easily * oops * fix * ours-no-visual testing * fix * interpret check * testing interpret approach * todo * fixes * test vila * setup experiments yaml! * revert bad changes
* setup to get results we missed * oops * revert yaml * fix tests * should improve the success rate of our method!
* run experiments * try something * change task dist * ready to run 5 seeds of experiments * run all baselines! * should be ready for merge * try vlm subselection baseline * testing * fix run config * new baseline is tested and ready * woops * woops * ready to go! * ready for review
…k distribution! (#1719) * Initial commit. * ready for experiments * change to 12 demos * align command with something that works * minior change * try this * new attempt pred_invention_vlm * try this * ready to go * single set of hyperparams that work! * yapf * good to go --------- Co-authored-by: Ashay Athalye <[email protected]>
* done! * fix linting
…put with another call to VLM (#1721) * Initial commit. * Clean up code a bit. * Get some checks to pass. * Update test. * Update tests. * Clean up code.
* should be ready for actual experiments! * ready for next run of parallel experiments * final set of experiments to run in parallel * fixes to prevent space from being eaten up * relaunch experiment that died due to space issues * try again on vila * try gemini * fatter_burger vila-fewshot * some changes to get interpret working * fix train tasks * fix changes * should be ready to go! * fixes
…ted baselines running in `kitchen` environment! (#1727) * adds set-of-marks labels for knobs * fix bugs in rendering * changes to get our approach working in kitchen * try to fix overfitting * try randomizing kitchen init state * try fdopt to speed up task planning * fixes vila * rerun new vila results on burger envs * fix config * rerun burger experiments * should be good to go! * rerun ours in kitchen * restore cluster utils * formatting * fixes * fix experiment yaml * linting
* fix vila * try combo_burger now * fix experiments * debug issue(s) in openai call * Make task solvable. * Break vila in combo-burger. * Add stuff to break vila-with-fewshot in all burger task types. * Uncomment stuff that was commented for debugging. * Get checks to pass outside of unittests. * Fix combo_burger test tasks. * Fix tests for burger env. * Run autoformatter. * Remove some debug comments. * Address comments on PR (mostly removing added spaces, etc.). * Revert changes made during debugging. --------- Co-authored-by: NishanthJKumar <[email protected]> Co-authored-by: Nishanth Kumar <[email protected]>
* start hacking on interpret * change yaml * hmmmmm - really not sure why interpret is failling... * found bug with interpret and fixed it! * update to run kitchen first * more stacks * fatter-burger * combo burger * good to go * try this * remove unnec change * simpler fix to interpret issue
* update fetch.urdf to help rendering * reset default torso lift height
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.