{
  "algorithm": "supermimic",
  "actor_input_normalization": "rsl_rl_empirical_real_next_obs",
  "task": "mug_hang",
  "seed": 1,
  "smoke": false,
  "reset_bank": "/mnt/data3/ZhouZhiqiang/supermimic_manipulation/banks/mug_hang.pt",
  "dataset": {
    "contract": {
      "task": "mug_hang",
      "robot": "arx5",
      "joint_order": [
        "joint1",
        "joint2",
        "joint3",
        "joint4",
        "joint5",
        "joint6",
        "joint7",
        "joint8"
      ],
      "fields": [
        "joint_pos",
        "joint_vel",
        "mug_pose",
        "mug_vel",
        "tree_pose"
      ],
      "frame": "environment_world",
      "pose": "xyz_wxyz",
      "joint_units": [
        "rad",
        "rad",
        "rad",
        "rad",
        "rad",
        "rad",
        "m",
        "m"
      ],
      "joint_velocity_units": [
        "rad/s",
        "rad/s",
        "rad/s",
        "rad/s",
        "rad/s",
        "rad/s",
        "m/s",
        "m/s"
      ],
      "position_unit": "m",
      "mug_velocity": "root_origin_local_linear_xyz_angular_xyz",
      "mug_velocity_units": [
        "m/s",
        "m/s",
        "m/s",
        "rad/s",
        "rad/s",
        "rad/s"
      ],
      "tree": "fixed_mocap_pose_constant_within_episode",
      "categories": [
        "ObjectAnywhereEEAnywhere",
        "ObjectRestingEEGrasped",
        "ObjectAirborneEEGrasped",
        "HandleThreadedEEGrasped"
      ]
    },
    "provenance": {
      "method": "fresh_rim_ik_and_osc_threading",
      "workspace": {
        "frame": "world",
        "xy_min": [
          0.2,
          -0.2
        ],
        "xy_max": [
          0.5,
          0.3
        ],
        "containment": "complete_object_geometry",
        "units": "m"
      },
      "workspace_geometry": "conservative_local_envelopes_of_all_visual_collision_geometry",
      "seed": 42,
      "proposal_config": {
        "parallel_chunk_size": 4,
        "x_range": [
          0.2,
          0.5
        ],
        "y_range": [
          -0.2,
          0.3
        ],
        "tree_y_range": [
          -0.2,
          0.3
        ],
        "airborne_z": [
          0.09,
          0.23
        ],
        "max_penetration": 0.001,
        "pre_thread_clearance": 0.04,
        "final_anchor_x": [
          0.04,
          0.065
        ],
        "witness_roll_samples": 129,
        "workspace": {
          "frame": "world",
          "xy_min": [
            0.2,
            -0.2
          ],
          "xy_max": [
            0.5,
            0.3
          ],
          "containment": "complete_object_geometry",
          "units": "m"
        },
        "grasp_point_mug_local": [
          0.0,
          -0.05,
          0.03
        ],
        "handle_anchor_mug_local": [
          0.0,
          0.03,
          0.0
        ],
        "task3": "outside_branch_tip_then_real_controller_threading_path",
        "orientation": "rim_closing_axis_roll; airborne_Haar_SO3; thread_axis_aligned",
        "jaw_turn_branches": [
          0.0,
          3.141592653589793
        ],
        "grasp_contact_search": "full_open_free_then_first_contiguous_contact_bracket",
        "witness_roll_order": "deterministic_uniform_grid_sorted_by_absolute_roll",
        "witness_scope": "independent_static_grasp_existence_not_transport_path",
        "resting_pose_count": 10000,
        "proposal_workers": 64,
        "proposal_seed_schedule": "seed_sequence_base_seed_and_ordered_chunk_index",
        "proposal_chunking": "fixed_size_ordered_chunks_dynamic_worker_assignment",
        "proposal_prefetch": "two_batches_bounded_ordered_results_except_postsettle_category",
        "proposal_budget_exhaustion": "return_partial_valid_rows_without_relaxing_predicates"
      },
      "proposal_attempts_per_state": 20000,
      "acceptance": {
        "max_penetration": 0.001,
        "stable_seconds": 0.5,
        "linear_speed_max": 0.03,
        "angular_speed_max": 0.3,
        "resting_linear_speed_max": 0.003,
        "resting_angular_speed_max": 0.05,
        "joint_speed_max": 2.0,
        "finger_upper_tolerance": 0.0004,
        "object_drift_max": 0.02,
        "minimum_thread_displacement": 0.04,
        "outside_tip_clearance": [
          0.015,
          0.03
        ],
        "outside_tip_transverse_jitter": 0.003,
        "outside_tip_transverse_max": 0.02
      },
      "num_envs": 1024,
      "physics_candidates": {
        "ObjectAnywhereEEAnywhere": 10240,
        "ObjectRestingEEGrasped": 12288,
        "ObjectAirborneEEGrasped": 12717,
        "HandleThreadedEEGrasped": 12288
      },
      "proposal_attempts": {
        "ObjectAnywhereEEAnywhere": 49992123,
        "ObjectRestingEEGrasped": 20987480,
        "ObjectAirborneEEGrasped": 350269887,
        "HandleThreadedEEGrasped": 15215118
      },
      "proposal_attempts_scope": "consumed_batches_only_excludes_unused_prefetch",
      "wall_time": {
        "initialization_seconds": 10.034658327698708,
        "categories": {
          "ObjectAnywhereEEAnywhere": {
            "proposal_seconds": 2746.244714386761,
            "physics_seconds": 440.1286013275385,
            "postsettle_seconds": 79.0680673494935,
            "total_seconds": 3311.5466208830476
          },
          "ObjectRestingEEGrasped": {
            "proposal_seconds": 1907.0405037105083,
            "physics_seconds": 67.6927042528987,
            "postsettle_seconds": 0.0001233816146850586,
            "total_seconds": 2039.4540275633335
          },
          "ObjectAirborneEEGrasped": {
            "proposal_seconds": 5513.964538797736,
            "physics_seconds": 93.07413939386606,
            "postsettle_seconds": 0.00019094347953796387,
            "total_seconds": 5695.0482866466045
          },
          "HandleThreadedEEGrasped": {
            "proposal_seconds": 1148.6602902188897,
            "physics_seconds": 306.5483791977167,
            "postsettle_seconds": 0.00014368444681167603,
            "total_seconds": 1518.3833962976933
          }
        },
        "total_seconds": 12629.966216474771,
        "proposal_includes_worker_startup": true,
        "proposal_seconds_scope": "consumer_wait_not_total_parallel_CPU_work",
        "resume_time_scope": "saved_checkpoint_segments_excludes_unsaved_interrupted_work",
        "category_total_scope": "completed_stage_timers_excludes_interrupted_stage_prefix"
      },
      "physics_dt": 0.0020833333333333333,
      "free_joint_damping": {
        "joint": "mug/mug_free",
        "dof_address": 8,
        "dof_count": 6,
        "linear": [
          0.0005,
          0.0005,
          0.0005,
          0.0005,
          0.0005,
          0.0005
        ],
        "quadratic": [
          0.0,
          0.0,
          0.0,
          0.0,
          0.0,
          0.0
        ],
        "cubic": [
          0.0,
          0.0,
          0.0,
          0.0,
          0.0,
          0.0
        ]
      },
      "control_dt": 0.05,
      "decimation": 24,
      "hold_steps": 20,
      "settle_steps": 200,
      "thread_steps": 60,
      "accepted_consecutive_steps": 10,
      "hold_command": "fixed_initial_ee_pose_via_relative_osc_action",
      "task3_path": "outside_tip_to_handle_engaged_by_osc_action_and_free_mug_dynamics",
      "controller": {
        "entity_name": "robot",
        "clip": null,
        "arm_joints": [
          "joint1",
          "joint2",
          "joint3",
          "joint4",
          "joint5",
          "joint6"
        ],
        "finger_joints": [
          "joint7",
          "joint8"
        ],
        "ee_body": "link6",
        "closed_position": 0.002,
        "open_position": 0.044,
        "effort_limit": 50.0,
        "inertia_solver": "pinv",
        "inertia_rcond": 1e-10,
        "uncouple_pos_ori": true,
        "position_scale": 0.02,
        "orientation_scale": 0.2,
        "motion_stiffness": [
          200.0,
          200.0,
          200.0,
          300.0,
          300.0,
          300.0
        ],
        "motion_damping_ratio": [
          3.0,
          3.0,
          3.0,
          1.0,
          1.0,
          1.0
        ],
        "input_limit": 12.0,
        "linear_velocity_limit": 1.5,
        "angular_velocity_limit": 8.0
      },
      "physics_randomization": false,
      "cuda_graph_mode": "plain",
      "nconmax": 512,
      "njmax": 2048,
      "contact_sensor_maxmatch": 128,
      "candidate_replay": {
        "all_exported_rows": true,
        "steps": 20,
        "required_consecutive_steps": 10,
        "saved_state": "replay_input_with_projected_fingers"
      },
      "collection_progress": {
        "enabled": false,
        "resumed": false,
        "boundary": null,
        "physical_continuation": "normal_reset_with_deterministic_batch_seed_not_inflight_simulation"
      },
      "finger_position_projection": {
        "joints": [
          "joint7",
          "joint8"
        ],
        "bounds_m": [
          [
            0.0,
            0.04399999976158142
          ],
          [
            0.0,
            0.04399999976158142
          ]
        ],
        "upper_overshoot_tolerance_m": 0.0004,
        "statistics": {
          "ObjectAnywhereEEAnywhere": {
            "changed_rows": 195,
            "max_correction_m": 6.332993507385254e-08
          },
          "ObjectRestingEEGrasped": {
            "changed_rows": 0,
            "max_correction_m": 0.0
          },
          "ObjectAirborneEEGrasped": {
            "changed_rows": 0,
            "max_correction_m": 0.0
          },
          "HandleThreadedEEGrasped": {
            "changed_rows": 0,
            "max_correction_m": 0.0
          }
        },
        "simulation_state_modified": false,
        "projected_contact_hold_revalidated": true
      }
    },
    "split": "train",
    "counts": {
      "ObjectAnywhereEEAnywhere": 10000,
      "ObjectRestingEEGrasped": 10000,
      "ObjectAirborneEEGrasped": 10000,
      "HandleThreadedEEGrasped": 10000
    }
  },
  "run": {
    "iterations": 9000,
    "buffer_steps": 1000,
    "warmup_steps": 64,
    "world_model_batches": 30,
    "world_model_only_iterations": 1000,
    "updates_per_sim_step": 0.4,
    "learning_rate_decay_iterations": 10000,
    "save_interval": 1000,
    "log_interval": 50,
    "action_noise": 0.1,
    "compile_world_model": true,
    "compile_policy_update": true,
    "pin_replay": false,
    "streaming_buffer_steps": 1000,
    "world_model_schedule": "epochs",
    "world_model_epochs": 4.0,
    "validation_fraction": 0.05,
    "validation_batches": 4,
    "validation_recent_steps": 100,
    "exploration_fraction": 0.0,
    "exploration_start_probability": 0.05,
    "exploration_steps": 10,
    "policy_update_interval": 1
  },
  "imagination_overrides": {},
  "world_model_validation_envs": 205,
  "agent": {
    "wm_hidden": [
      2048,
      2048,
      2048,
      2048,
      2048
    ],
    "policy_hidden": [
      2048,
      1024,
      512,
      256
    ],
    "critic_hidden": [
      256,
      128
    ],
    "wm_window": 8,
    "horizon": 16,
    "wm_batch_size": 2048,
    "policy_batch_size": 4096,
    "learning_rate": 0.0002,
    "gamma": 0.99,
    "lambda_return": 0.95,
    "critic_epochs": 5,
    "critic_minibatches": 4,
    "max_grad_norm": 1.0,
    "policy_grad_value_clip": 0.1,
    "normalizer_max_count": 200000000.0,
    "amp": "bf16",
    "tf32": true,
    "wm_ensemble_size": 1,
    "uncertainty_penalty": 0.0
  },
  "device": "cuda:0",
  "device_name": "NVIDIA A100-SXM4-80GB",
  "torch": "2.9.1+cu128",
  "cuda": "12.8",
  "precision": {
    "parameters_normalizers_state_reward": "float32",
    "mlp_autocast": "bf16",
    "controller_ik": "float64",
    "matmul_allow_tf32": true,
    "cudnn_allow_tf32": true
  },
  "torch_compile": true,
  "torch_compile_scope": "world_model_rollout_loss",
  "torch_compile_options": {
    "fullgraph": true,
    "dynamic": false,
    "mode": "default"
  },
  "torch_compile_policy_update": true,
  "logger": "tensorboard",
  "cuda_graph_mode": "plain",
  "timeout_bootstrap": false,
  "resume_is_exact": false,
  "resume_checkpoint": "/mnt/data3/ZhouZhiqiang/supermimic_manipulation/logs/supermimic/mug_hang_s1_20261026/model_001000.pt",
  "evaluation": "separate complete-episode workflow (clab eval policy)",
  "limitations": [
    "Bank controller match is not a physical hold/capacity certificate",
    "World model is a reduced stochastic surrogate, not an exact MuJoCo state",
    "Imagination reward/success is analytic from endpoint states, not a learned reward head",
    "Endpoint imagination reward/success is approximate and need not match native contacts",
    "Real collection and evaluation retain native rewards and terminal conditions"
  ],
  "reward_mode": "imagined_endpoint_handle_geometry_speed_release_shaped",
  "reward_matches_native": false,
  "cuda_graph_runtime": {
    "enabled": true,
    "graph_conditional": false,
    "step_graph_created": true,
    "forward_graph_created": true,
    "reset_graph_created": true
  }
}