Learn to predict, then plan

CPU · Python 3.12.13 · 4.23 s · 2026-10-22 08:56:13 UTC+8

Saved run results

test_state_mse	0.032912276685237885
zero_delta_state_mse	0.0724581927061081
open_loop_mse_at_29	21.297061988245694

{'train_shapes': [[4200, 2], [4200, 1], [4200, 3]], 'test_shapes': [[1200, 2], [1200, 1], [1200, 3]], 'split': 'whole trajectories'}

{'zero_delta_state_mse': 0.0724581927061081, 'learned_state_mse': 0.032912276685237885}

Random true returns: [-804.9, -2626.5, -118.5, -2516.4, -80.6, -142.8]
Zero torque true returns: [-250.0, -250.0, -1010.4, -1010.4, -330.0, -330.0]

Learned MPC true returns: [-29.0, -23.7, -19.2, -17.7, -5.8, -4.6]

Optional exercise: fill answer and enable RUN_EXERCISE.

metrics.json

{
  "test_state_mse": 0.032912276685237885,
  "zero_delta_state_mse": 0.0724581927061081,
  "open_loop_mse_at_29": 21.297061988245694,
  "true_returns": {
    "Random": [
      -804.86215302255,
      -2626.5262360572815,
      -118.54748625401407,
      -2516.35036611557,
      -80.57731036795303,
      -142.7590591609478
    ],
    "Zero torque": [
      -250.0,
      -250.0,
      -1010.3533520698547,
      -1010.3533520698547,
      -329.9672239422798,
      -329.9672239422798
    ],
    "Learned MPC": [
      -29.049718314316124,
      -23.740234406664968,
      -19.19470847863704,
      -17.72059811418876,
      -5.8373159766197205,
      -4.604384808801115
    ]
  }
}