Minor Errors - #97
Open
rojas70 wants to merge 290 commits into
Open
Conversation
…into fractal_replay_buffer
…gisticsLab/serl into fractal_replay_buffer
Improved documentation will be desirable. Can also include links to notion in the documentation for more in-depth treatment.
- Modularized methods - Introduced PD-controller vs just P-controller. - Kp, Kv gains set at 16,20.
Removed weld operations they are not present currently in franka_sim.
- testing.sh and tmux_script_launch.sh should be used together - should be made executable and called ./ - still incomplete and with a bug showing up in actor/learner py scripts. but close
- Improved index handling for environments - Updated how env.step() returns done based on the reward definition in this reach env. Used to always be False. - Path setting improvements to save .npz file. - Created a load_demo_test.py to assert written data can be read and parsed by episode and time step.
…into fractal_replay_buffer
…async_sac_state. - franka_reach_demo_script.py: now keeps count of total transitions - demoHandling.py: constructor checks for directories/files and loads all data to member - method insert_data_to_buffer pushes releveant transition data to QueuedDataStore for actor
…hance tests for fractal_branch and time_split methods
…tely use demo data in the replay buffer. - Script was producing actions that had dimensions in the 100s. After clipping, values were always \pm1. This would not allow the neural network to learn well from these numbers. An ACTION_MAX value was set to 10 to produce more realistic actions. TODO: analyze natural action dimensions in SERL and approximate as closely as possible. - In the franka_reach_demo_script, updated error_thrshold used to get gripper closer to object. Currently set at 0.01m. - This changes the reward values produced. Updated panda_reach_gym_env step function such that when the rewards is greater than some threshold we return a done that is True. Updated async_sac_state - increase size of actor QueuedDataStore based on the total number of transitions available in the demos. - set training_starts to 0 since we have good demo data to start with.
…ed testing can be done
…ab/serl into demos
…erl into object_relocation
Fix/documentation changes
Updated contributions section to clarify the Fractal SERL Framework and SymmGrid.
Updated images in the README to reflect new training visuals.
Updated image formatting and citations in README.
Updated the README to include details about homographies and recent updates.
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
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.
Commit 422d207 fixes a bug in the way the Mujoco Renderer is instantiated so that there is no crush when calling. test_gym_env_human.py.
Commit 7e23bbd includes a critic_actor_ratio that is missing when agent.high_utd_ratio is called.
Others...