anarcho_av---multi-agent-rl-for-av-traffic-clearance's People
Forkers
timefly-1989anarcho_av---multi-agent-rl-for-av-traffic-clearance's Issues
Reference papers
Hello
Thank you for sharing your code. Could you upload your paper? I tried to download but I could not find
In env.py, the function get_follow_speed_by_id, there is a mistake
At the beginning of the function
leader_data = traci.vehicle.getLeader(self.emer.ID)
the second line should get sent the agent vehicle Id not the emergency vehicle ID
no_acc action allows acceleration
The current implementation of the no_acc (no acceleration) action, allows acceleration. This happens because the default used behaviour of SUMO is to apply maximum possible acceleration, and our implementation of no_acc was to simply send nothing to SUMO.
Our edits should:
- make sure no_acc is applied as: current velocity is maintained in singleAgent.applyAction function.
- make sure the feasibility check in env.get_feasible_actions checks for the feasibility of maintaining the current velocity. My current belief is that: no_acc will always be feasible, since:
- Forward feasibility will need to check if the current velocity is greater than vsafe (safe velocity for next step, retrieved by traci.getFollowSpeed()), if so, no_acc action should be addressed as infeasible.
- It does not require |deceleration| > |max_deceleration| --> So it is always backward feasible
- Integarete this with are_we_ok function created by @AlaaHesham1996 and check if last step's velocity was maintained for our agent.
Footnote: One thing that we must maintain in our system to avoid breaking it down, though.. is: there should always be at least one always-feasible action. This is to guarantee that the agent will always be able to choose an action per loop, as assumed. Currently, the only always-feasible action is decelerate.
Condition for checking reward on maximum velocity is faulty
Calculating the step reward in the Single Agent (Version 1.3), here has something wrong to it.:
The reward, in general, is a linear function of the acceleration. However, if the EV (ambulance) was already at maximum velocity and maintained maximum velocity for the next step, then we will see its acceleration as zero although no agent affected its acceleration or slowed it down. Our logic was then to give the agent the maximum reward nonetheless (i.e. equivalent to maximum acceleration.
This was --however-- translated in a somewhat wrong way to code. We see that the condition for the reward is:
if ( abs(amb_last_velocity-self.emer.max_speed) <= 1e-10 ): #i.e. amb_last_velocity==self.emer.max_speed
Which only reflects the first part (ambulance was at maximum velocity), without "and maintained maximum velocity for the next step". The corrected code should look something like:
if ( abs(amb_last_velocity-self.emer.max_speed) <= 1e-10 and abs(self.emer.spd-self.emer.max_speed) <= 1e-10 ):
I did not edit it so that we would get a chance to re-train the agent before updating the code, and see if that affects anything.
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. ๐๐๐
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google โค๏ธ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.