Data-Driven Multi-Agent Vehicle Routing in a Congested City

Round 1

Reviewer 1 Report

The work concerns an interesting topic. It has a very popular scientific character. Rather not very scientific. But its level is acceptable.

Figure 2 is not legible.

I have two comments on this work:
1. The described experimental studies are very general. In my opinion, the detailed algorithm of the obtained results is missing. There are no results (there is no possibility) that would allow the verification of the obtained conclusions.
2. The work does not formally correspond to the publisher's template.

Third note (written a bit "forcibly"): the article is based on relatively old literature. In my opinion, the latest knowledge on this subject has been partially overlooked.

Author Response

Thank you very much for the review and valuable comments!

1. We have replaced Figure 2 with a larger screenshot with a proper format and put a sentence as the new title of the figure in the paper.

2. To better explain the simulation process, we included some additional text regarding the SUMO simulator on page 8 under section 4.1.  

3. The whole paper has been re-formatted to correspond to the publisher’s template.

4. One paragraph in the literature part has been added to include new reviews on the recent research in this topic. Three new references have been added for this.

Reviewer 2 Report

The paper proposes a multi-agent routing methodology the objectives of improving the navigation in a traffic congested city. It is based on using the A* algorithm, a centralized real-time traffic information system (TIS) that  returns two average travel times for each road segment requested, adjusted for  the time at which the agent estimates it will reach the segment (1. the long-term average - which is an average of all vehicle travel times on the given road segment for all available routing episodes; 2. the short-term average, comprised of the average travel time for all vehicles in  the certain number of most recent routing episodes), the SAW formula used by the agent to apply the averages to  each potential route. The route is selected with the fastest estimated time and the routing method can adapt to changes in congestion (previous travel time data are used to develop the estimates).

Strengths:

1. By combining the data the driver acquires through experience driving a route with the data that is collected by all road users, the fastest route in a congested road network is determined with less exploration than might be necessary in other situations
2. by constructing a new reinforcement learning based approach to teaching each driver which travel data best matches the current congestion, the proposed method is adaptable to changes in the congestion problem
3. the use of the multi-agent simulation for demonstrating that the drivers can reach an equilibrium point that approaches the system optimum with being directed through a control mechanism

Weakness:

1. The formulas must be re-written to be more clear (eq.1, 2, 3 ...)
2. In Figure 9 times at equilibrium does not appear; it must be explained more clear the reason?
3. The references could be chosen to be more recent

Author Response

Thank you very much for the review and valuable comments!

1. Formulas 1 and 2 have been changed for brevity, but we some difficulty shortening formula 3.
2. The issue with figure 9 not displaying the equilibrium values is due to no equilibrium being found for Re-routes with TIS. This is noted in the text on page 12 under ‘User Equilibrium Points and Minimum Re-routes’.