Table 6 Overall performance comparison across algorithms. DQN-based methods report mean ± standard deviation over 12 independent runs with 20 test episodes per run. T-APF reports mean ± standard deviation over 20 test episodes due to stochastic escape behavior. E-APF is deterministic and reports single-run values.
Algorithm | Success rate (%) | Path length (m) | Decision steps | Energy consumption |
|---|---|---|---|---|
T-APF | 95.0 | \(348.18 \pm 63.35\) | \(92.50 \pm 16.26\) | \(116.99 \pm 30.37\) |
E-APF | 100.0 | 270.09 | 47.00 | 53.67 |
DQN | 100.00 | \(294.69 \pm 20.86\) | \(50.85 \pm 6.17\) | \(76.04 \pm 13.30\) |
DQN-EG | 100.00 | \(307.91 \pm 5.43\) | \(47.05 \pm 1.47\) | \(61.81 \pm 3.37\) |
APF-DQN-NOC | 100.00 | \(312.35 \pm 11.01\) | \(55.55 \pm 3.41\) | \(79.53 \pm 7.46\) |
APF-DQN | 100.00 | \(262.85 \pm 4.18\) | \(41.30 \pm 0.84\) | \(49.55 \pm 2.05\) |
