release/0.2.2

What's Changed

• multi-constrained sbsrl safe cartpole experiment by @lucavignola in https://github.com/yardenas/safe-learning/pull/235
• Add rccar rae by @yardenas in https://github.com/yardenas/safe-learning/pull/236
• uncertainty critic and real data by @lucavignola in https://github.com/yardenas/safe-learning/pull/237
• Add tasks for nonepisodic humanoid and implementation via terminations by @yardenas in https://github.com/yardenas/safe-learning/pull/238
• Match performance sac by @yardenas in https://github.com/yardenas/safe-learning/pull/239
• Sbsrl priors g2g by @lucavignola in https://github.com/yardenas/safe-learning/pull/240
• Normalize disagreement by @lucavignola in https://github.com/yardenas/safe-learning/pull/241
• sbsrl offline by flipping the uncertainty constraint by @lucavignola in https://github.com/yardenas/safe-learning/pull/242
• Remove some files by @yardenas in https://github.com/yardenas/safe-learning/pull/243
• Load the behavior action when training the actor critic by @yardenas in https://github.com/yardenas/safe-learning/pull/244
• Terminate humanoid by @yardenas in https://github.com/yardenas/safe-learning/pull/246
• sbsrl new sgd update for SAC by @lucavignola in https://github.com/yardenas/safe-learning/pull/245
• compute sgd step in parallel over ensemble dimension by @lucavignola in https://github.com/yardenas/safe-learning/pull/247
• sbsrl_offline - sooper compatibility by @lucavignola in https://github.com/yardenas/safe-learning/pull/248
• Migrate to UV by @yardenas in https://github.com/yardenas/safe-learning/pull/250

UV migration, results for online RL done

Full Changelog: yardenas/safe-learning@release/0.2.1...release/0.2.2

release/0.2.1

What's Changed

• Unsafe rccar by @yardenas in https://github.com/yardenas/safe-learning/pull/230
• sbsrl initial commit by @lucavignola in https://github.com/yardenas/safe-learning/pull/229
• Implementation of Ensemble critic by @lucavignola in https://github.com/yardenas/safe-learning/pull/232
• actor loss adapted to use critic predictions(mean) by @lucavignola in https://github.com/yardenas/safe-learning/pull/234
• Getup cmdp by @yardenas in https://github.com/yardenas/safe-learning/pull/233
• Add experiment files for online on Go2 by @yardenas in https://github.com/yardenas/safe-learning/pull/231

Go1-2 online RL experiments

New Contributors

• @lucavignola made their first contribution in https://github.com/yardenas/safe-learning/pull/229

Full Changelog: yardenas/safe-learning@release/0.2.0...release/0.2.1

release/0.2.0

What's Changed

• Refactor out the sooper on policy training step fn by @yardenas in https://github.com/yardenas/safe-learning/pull/220
• Remove unused experiments by @yardenas in https://github.com/yardenas/safe-learning/pull/221
• Implementation of actsafe by @yardenas in https://github.com/yardenas/safe-learning/pull/223
• Nonepisodic Learning by @yardenas in https://github.com/yardenas/safe-learning/pull/222
• Improvements to franka experiment by @yardenas in https://github.com/yardenas/safe-learning/pull/224
• Fix data loading for RAE by @yardenas in https://github.com/yardenas/safe-learning/pull/225
• Parameter for pessimistic q by @yardenas in https://github.com/yardenas/safe-learning/pull/226
• Humanoid standup by @yardenas in https://github.com/yardenas/safe-learning/pull/227
• Nonepisodic humanoid by @yardenas in https://github.com/yardenas/safe-learning/pull/228

Full Changelog: yardenas/safe-learning@release/0.1.25...release/0.2.0

release/0.1.25

What's Changed

• Add fingers to Franka by @yardenas in https://github.com/yardenas/safe-learning/pull/211
• Rccar hardware baseline by @yardenas in https://github.com/yardenas/safe-learning/pull/212
• Use red box by @yardenas in https://github.com/yardenas/safe-learning/pull/213
• add lbsgd penalizer and example config for cartpole swingup by @yardenas in https://github.com/yardenas/safe-learning/pull/216
• Add implementation of pure exploration by @yardenas in https://github.com/yardenas/safe-learning/pull/217
• Second attempt on offline rccar + hardware by @yardenas in https://github.com/yardenas/safe-learning/pull/214
• minor changes to log barriers implementation and finish tuning by @acpoll in https://github.com/yardenas/safe-learning/pull/218
• Walker offline by @ManuelWendl in https://github.com/yardenas/safe-learning/pull/219

Full Changelog: yardenas/safe-learning@release/0.1.24...release/0.1.25

release/0.1.24

What's Changed

• Implementation of sooper with vision by @yardenas in #205
• Add breaking controller by @yardenas in #207
• Lagrangian baselines by @acpoll in #210
• Sooper go-to-goal v2 by @ManuelWendl in #209

Final parameters for SOOPER paper

Full Changelog: release/0.1.23...release/0.1.24

release/0.1.23

What's Changed

• MBPO + Vision by @yardenas in #200
• SOOPER + Vision by @yardenas in #202
• Fix bugs in sooper by @yardenas in #204
• Updates to Franka Online by @yardenas in #203
• Variable goal size for go-to-goal environment by @acpoll in #206

Online learning on franka works

Full Changelog: release/0.1.22...release/0.1.23

release/0.1.22

What's Changed

• Load auxiliaries by @yardenas in #189
• Add PointGoal1 by @yardenas in #190
• Sooper humanoid by @ManuelWendl in #191
• Take other fix by @yardenas in #192
• Handle safe penalizer without sooper by @yardenas in #193
• Take other fix by @yardenas in #194
• Fix normalizer in offline by @yardenas in #195
• Sooper gtg easy offline by @ManuelWendl in #196
• Add offline data collection by @yardenas in #197
• Logging updates by @yardenas in #198
• SPiDR from Vision by @yardenas in #177
• Fix MBPO vision wrapper by @yardenas in #199
• Sooper go to goal v1 improved version by @ManuelWendl in #201

Offline on PointGoal1

Full Changelog: release/0.1.21...release/0.1.22

release/0.1.21

What's Changed

• Observation wrapper safety gym by @ManuelWendl in #185
• Observation wrapper walker by @ManuelWendl in #187
• Preprocess RBG images by @yardenas in #172
• Sooper walker walk by @ManuelWendl in #188

Walker baseline and franka arm sim-to-real

Full Changelog: release/0.1.20...release/0.1.21

release/0.1.20

What's Changed

• Sooper cartpoleswingup by @ManuelWendl in #169
• Franka hardware debug image by @yardenas in #171
• Get offline training working by @ManuelWendl in #173
• Collect offline data by @yardenas in #174
• Offline config by @ManuelWendl in #175
• Sooper rccar by @ManuelWendl in #176
• MBPO no pessimistic sample by @yardenas in #178
• if case dict by @yardenas in #179
• Minor utility things by @acpoll in #180
• Eval policy by @acpoll in #182
• Fix wrapping of eval env in safety gym by @ManuelWendl in #183
• Rccar Hardware Offline Training by @ManuelWendl in #184

Offline to online works

Full Changelog: release/0.1.19...release/0.1.20

release/0.1.19

What's Changed

• Franka hardware training by @yardenas in #165
• Hardware on RaceCar by @yardenas in #170

Full Changelog: release/0.1.18...release/0.1.19