FedE2E: One-Shot Federated Learning for End-to-End Autonomous Driving with Closed-Loop Evaluation
Organizations: Computer Vision Center (CVC), Univ. Aut`onoma de Barcelona (UAB), Barcelona, Spain
Abstract
Federated learning enables collaborative model training without sharing datasets. However, existing federated learning methods for autonomous driving typically require multi-round communication and lack closed-loop evaluation, while one-shot methods mainly focus on perception tasks. To address these limitations, we propose \method, a one-shot federated learning method for end-to-end autonomous driving. It comprises multi-client calibration, anchor-based feature alignment, and token-wise consensus distillation to distill knowledge from heterogeneous models. Closed-loop evaluation on Bench2Drive demonstrates the effectiveness of \method in both one-shot and multi-round federated learning. In the one-shot setting, it achieves a Driving Score of 55.85, which is 4.07 points higher than prior work. In the multi-round setting, it reaches 60.35 after five communication rounds, the best performance among all compared methods under the same number of communication rounds.
Figures & tables
| Method | DS | SR (%) | RC (%) | IP |
| FedAvg [ 3 ] | 7.05 0.12 | 0.00 0.00 | 8.45 0.16 | 0.93 0.00 |
| FedRDN [ 40 ] | 14.99 0.09 | 0.00 0.00 | 44.75 0.28 | 0.36 0.00 |
| FedProx [ 20 ] | 26.32 0.35 | 1.82 0.00 | 58.39 0.59 | 0.47 0.00 |
| FedDF [ 21 ] | 51.78 0.50 | 20.61 0.57 | 69.73 0.06 | 0.73 0.01 |
| FedE2E (Ours) | 55.85 0.23 | 24.70 0.93 | 71.51 0.56 | 0.78 0.01 |
| Centralized | 63.30 0.63 | 29.24 0.57 | 80.08 0.69 | 0.78 0.01 |
| Model | DS | SR (%) | RC (%) | IP |
| Town03 | 41.88 0.64 | 14.55 1.62 | 58.33 0.63 | 0.70 0.00 |
| Town12 | 46.89 0.62 | 14.24 0.93 | 69.04 0.88 | 0.68 0.00 |
| Town13 | 50.01 0.72 | 18.79 1.55 | 69.52 0.61 | 0.71 0.00 |
| Town15 | 29.84 0.97 | 4.24 0.21 | 52.12 0.97 | 0.56 0.01 |
| FedE2E (Ours) | 55.85 0.23 | 24.70 0.93 | 71.51 0.56 | 0.78 0.01 |
| Centralized | 63.30 0.63 | 29.24 0.57 | 80.08 0.69 | 0.78 0.01 |
| Method | Overtaking | Merging | Emergency Brake | Give Way | Traffic Signs | Ability Avg. |
| FedAvg | 0.00 0.00 | 0.00 0.00 | 0.00 0.00 | 0.00 0.00 | 4.91 0.25 | 0.98 0.05 |
| FedProx | 0.00 0.00 | 2.50 0.00 | 3.33 0.00 | 0.00 0.00 | 15.09 0.25 | 4.18 0.05 |
| FedRDN | 0.00 0.00 | 0.00 0.00 | 0.00 0.00 | 0.00 0.00 | 30.35 0.66 | 6.07 0.13 |
| FedDF | 18.52 4.19 | 19.58 1.18 | 23.89 0.79 | 56.67 9.43 | 17.19 1.79 | 27.17 1.24 |
| FedE2E (Ours) | 28.15 1.05 | 23.75 1.77 | 28.33 2.36 | 46.67 17.00 | 24.04 1.38 | 30.19 3.10 |
| Centralized | 33.33 1.81 | 28.75 0.00 | 33.33 2.36 | 36.67 4.71 | 27.19 1.31 | 31.86 1.28 |
| Method | Overtaking | Merging | Emergency Brake | Give Way | Traffic Signs | Ability Avg. |
| Town03 | 9.63 1.05 | 12.08 1.18 | 18.89 4.16 | 63.33 4.71 | 16.14 1.08 | 24.02 1.78 |
| Town12 | 17.04 5.24 | 14.58 1.56 | 13.33 1.36 | 30.00 8.16 | 20.00 2.27 | 18.99 2.50 |
| Town13 | 15.56 4.80 | 19.17 0.59 | 21.11 1.57 | 43.33 4.71 | 17.19 1.79 | 23.27 1.40 |
| Town15 | 1.48 1.05 | 2.50 0.00 | 6.11 0.79 | 16.67 4.71 | 11.75 0.66 | 7.70 0.95 |
| FedE2E (Ours) | 28.15 1.05 | 23.75 1.77 | 28.33 2.36 | 46.67 17.00 | 24.04 1.38 | 30.19 3.10 |
| Centralized | 33.33 1.81 | 28.75 0.00 | 33.33 2.36 | 36.67 4.71 | 27.19 1.31 | 31.86 1.28 |
| Method | DS | SR (%) | RC (%) | IP |
| FedAvg | 58.54 1.40 | 25.00 1.34 | 78.20 1.14 | 0.74 0.01 |
| FedProx | 26.86 0.64 | 2.58 0.57 | 59.65 0.36 | 0.46 0.01 |
| FedRDN | 45.94 0.80 | 14.24 1.30 | 72.13 0.87 | 0.63 0.00 |
| FedDF | 58.77 0.56 | 26.67 0.21 | 74.79 0.55 | 0.77 0.01 |
| FedE2E (ours) | 60.35 0.62 | 26.97 0.77 | 79.52 0.66 | 0.75 0.01 |
| Round | DS | SR (%) | RC (%) | IP |
| 1 | 55.85 0.23 | 24.70 0.93 | 71.51 0.56 | 0.78 0.01 |
| 2 | 56.84 0.32 | 21.21 1.50 | 77.22 0.05 | 0.74 0.01 |
| 3 | 59.58 0.38 | 25.61 1.67 | 78.96 0.62 | 0.75 0.00 |
| 4 | 58.32 0.80 | 25.15 1.83 | 78.09 0.17 | 0.74 0.01 |
| 5 | 60.35 0.62 | 26.97 0.77 | 79.52 0.66 | 0.75 0.01 |
| Distillation | Method | DS | SR (%) | RC (%) | IP |
| Town03 | FedE2E | 57.47 0.71 | 24.09 1.48 | 71.31 0.54 | 0.80 0.00 |
| FedDF | 49.42 0.68 | 19.09 0.37 | 65.77 0.06 | 0.73 0.01 | |
| Town12 | FedE2E | 55.85 0.23 | 24.70 0.93 | 71.51 0.56 | 0.78 0.01 |
| FedDF | 51.78 0.50 | 20.61 0.57 | 69.73 0.06 | 0.73 0.01 | |
| Town13 | FedE2E | 54.12 0.85 | 20.15 1.13 | 66.76 0.58 | 0.82 0.01 |
| FedDF | 49.47 0.40 | 17.12 0.43 | 68.58 0.70 | 0.70 0.00 |
| MCC | ABA | TCD | DS | SR (%) | RC (%) | IP |
| – | ✓ | ✓ | 52.82 0.48 | 20.45 0.74 | 70.49 0.61 | 0.75 0.00 |
| ✓ | – | ✓ | 50.80 0.75 | 17.88 2.39 | 69.22 0.27 | 0.75 0.01 |
| ✓ | ✓ | – | 54.46 1.25 | 23.48 1.19 | 70.58 0.77 | 0.77 0.01 |
| ✓ | ✓ | ✓ | 55.85 0.23 | 24.70 0.93 | 71.51 0.56 | 0.78 0.01 |
Federated Learning
Federated learning trains shared models across data holders without requiring them to pool their raw data.
See all 1,037 papers