feat: config setup for lazy bums

configs/best_config_so_far.yaml

@@ -0,0 +1,95 @@
+architecture:
+  critic:
+    activation: tanh
+    hidden_dims: []
+  feature_extractor:
+    activation: tanh
+    hidden_dims:
+    - 300
+    - 300
+    - 300
+  message_passing_steps: null
+  motor:
+    activation: tanh
+    hidden_dims: []
+  name: centralized
+  propagator: null
+  sensor:
+    activation: tanh
+    hidden_dims:
+    - 300
+    - 300
+    - 300
+  topology_type: null
+arena:
+  attach_target: true
+  sand_ground_color: true
+  size:
+  - 10.0
+  - 5.0
+  wall_height: 1.5
+  wall_thickness: 0.1
+environment:
+  camera_ids:
+  - 0
+  - 1
+  joint_randomization_noise_scale: 0.0
+  light_perlin_noise_scale: 0
+  num_physics_steps_per_control_step: 10
+  render_size:
+  - 480
+  - 640
+  simulation_time: 5000.0
+  target_distance: 3.0
+  task: !!python/object/apply:brittle_star_project.environment.env_types.Task
+  - directed_locomotion
+  time_scale: 2
+experiment:
+  base_run_dir: runs
+  cuda: true
+  debug_sanity: false
+  exp_name: lr lowered
+  seed: 42
+  torch_deterministic: true
+logging:
+  capture_video: false
+  checkpoint_frequency: 100
+  hf_entity: ''
+  save_checkpoints: true
+  save_model: true
+  track: true
+  upload_checkpoints: false
+  upload_final_model: true
+  wandb_entity: SEL3-2026-Groep-4
+  wandb_project_name: Reducing randomness
+morphology:
+  segments_per_arm:
+  - 4
+  - 0
+  - 4
+  - 0
+  - 0
+  use_p_control: true
+  use_torque_control: false
+ppo:
+  anneal_lr: true
+  clip_coef: 0.2
+  clip_vloss: true
+  ent_coef: 0.001
+  gae_lambda: 0.95
+  gamma: 0.99
+  learning_rate: 0.0001
+  max_grad_norm: 0.5
+  norm_adv: true
+  num_envs: 32
+  num_minibatches: 32
+  num_steps: 64
+  target_kl: 0.02
+  total_timesteps: 1228800
+  update_epochs: 4
+  vf_coef: 1.0
+simulation:
+  backend: !!python/object/apply:brittle_star_project.environment.env_types.Backend
+  - MJX
+  model_path: null
+  model_type: random

configs/environment/dir_loc_further.yaml

@@ -0,0 +1,2 @@
+simulation_time: 50000.0
+target_distance: 3.0

configs/experiment/long_2arm.yaml

@@ -0,0 +1,6 @@
+# Testing chicken dinner 4 but further distance.
+
+exp_name: "long2arm"
+seed: 123
+torch_deterministic: true
+cuda: true

configs/logging/pushing_long.yaml

@@ -0,0 +1,10 @@
+capture_video: false
+checkpoint_frequency: 100
+hf_entity: ''
+save_checkpoints: true
+save_model: true
+track: true
+upload_checkpoints: false
+upload_final_model: true
+wandb_entity: "SEL3-2026-Groep-4"
+wandb_project_name: "Pushing our best found config"

configs/ppo/chickendinnerwinner.yaml

@@ -0,0 +1,16 @@
+anneal_lr: true
+clip_coef: 0.2
+clip_vloss: true
+ent_coef: 0.001
+gae_lambda: 0.95
+gamma: 0.99
+learning_rate: 0.0001
+max_grad_norm: 0.5
+norm_adv: true
+num_envs: 32
+num_minibatches: 32
+num_steps: 64
+target_kl: 0.02
+total_timesteps: 12288000
+update_epochs: 4
+vf_coef: 1.0

scripts/hpc/train.pbs

@@ -66,8 +66,11 @@ fi
 # Run training using Hydra overrides
 python scripts/train.py \
     hydra.run.dir="$SCRATCH_RUNDIR" \
-    ppo=stable \
+    environment = dir_loc_further \
-    logging=hpc
+    experiment = long_2arm \
+    ppo=chickendinnerwinner \
+    logging=pushing_long
+
 
 echo ">>> Staging out results to $DATA_RUNDIR..."
 cp -r "$SCRATCH_RUNDIR/." "$DATA_RUNDIR/"