Page 66 - Read Online
P. 66

Page 164                        Boin et al. Intell Robot 2022;2(2):145­67  I http://dx.doi.org/10.20517/ir.2022.11



























































                                   (a) No-FRL                             (b) Intra-FRLWA

               Figure 11. Results for a specific 60s test episode using the 5 vehicle 1 platoon environment trained using no-FRL (Figure 11a), and with
               Intra-FRLWA (Figure 11b).


               4. CONCLUSION
               In this paper, we have formulated an AV platooning problem and successfully applied FRL in a variety of meth-
               ods to AV platooning. In addition, we proposed new approaches for applying FRL to AV platoons: Inter-FRL
               and Intra-FRL. By comparing FRL performance with both gradient and weight averaging in the AV platoon-
               ing scenario, it has been shown that weight averaging was the optimal aggregation method regardless of using
               Inter-FRL or Intra-FRL. Furthermore, it was found that the Intra-FRLWA strategy was most advantageous for
               applying FRL to AV platooning. Finally, it was proven that applying Intra-FRLWA to AV platoons up to 5 vehi-
               cles in length provided large performance advantages during and after training when compared to AV platoons
   61   62   63   64   65   66   67   68   69   70   71