Table 4 Evaluation metrics for NMPC using the estimated model parameters from the DRL agent in the state prediction of the NMPC, compared to the actual model parameters used for the state prediction.

From: Digital twin syncing for autonomous surface vessels using reinforcement learning and nonlinear model predictive control

Metric

DRL agent

Actual model parameters

Episodes

100.00

100.00

Avg. cross-track error

2.67

3.98

Avg. std cross-track error

3.38

3.64

Avg. max cross-track error

17.10

18.02

Avg. timesteps

203.71

203.92

Goals reached

97%

100.00%

Infeasible solutions

0.0

0.0

Avg. heading error

0.00

-0.01

Avg. mean heading error

0.00

-0.01

Avg. std heading error

0.09

0.08

Avg. max heading error

0.42

0.39

Avg. surge speed error

0.13

0.13

Avg. mean surge speed error

0.11

0.12

Avg. std surge speed error

0.39

0.39

Avg. max surge speed error

1.70

1.69