Tick function for Reinforcement Learning

[kraken24]

Hello Everyone,

I am using Carla 0.9.4 for developing reinforcement learning algorithm for autonomous driving.

I have 2 queries:

1. I am confused between world.tick, world.on_tick and world.wait_for_tick functions. I do not understand which function to use when?

2. I want simulator to perform certain step only when I give signals and it should stop when I train my neural network.

I have checked synchronous_mode.py but i still get error that: failed to connect to newly created map

[nsubiron]

Hi @kraken24,

First of all, 0.9.4 has several issues with synchronous mode that have already being addressed in master, among them dead-locks when changing the map. Also take a look at #1803 as it may have some further fixes.

The thing you need to know is that every time the simulator does an update (advances the simulation one step), it sends a message to the clients that are connected. We also call this update a "tick".

• world.wait_for_tick() blocks until a new tick is received from the simulator. This function is not very reliable as you may skip ticks, and you can easily end up in a dead-lock if using synchronous mode.
• world.on_tick(callback) registers a callback (function or function object) to be called every time a new tick is received, the callback is executed asynchronously in the background.
• world.tick() tells the simulator to perform a tick (i.e. advance one step). This only has effect in synchronous mode.

So if you enable synchronous mode, the simulator is going to halt until you call world.tick().

[kraken24]

thank you @nsubiron

Now i am able to work with synchronous mode.

But I have 1 more problem, after spawining the vehicle and setting a fix velocity as initial setting,

1. if I use set_velocity(vehicle.get_velocity()), it works perfectly and I get constant velocity of vehicle
2. if I use vehicle.apply_control(throttle=1) then vehicle decelerates few seconds and then accelerates.

is it because I use only 10 fps?

[nsubiron]

set_velocity(...) bypasses the vehicle model, it just applies the velocity directly in the physics engine, so you get the net velocity you want. On the other hand, apply_control(...) uses the vehicle model to calculate the final velocity; i.e. takes into account gear shifting, engine RPM, etc. Probably the vehicle decelerates because is shifting gears.

[kraken24]

Thanks @nsubiron

Is there any documentation that explains relationship between carla vehicle control parameters and vehicle parameters. for example variation of rate of acceleration as i change throttle values. (all other parameters are kept constant)

[nsubiron]

Not really, we just pass these values to the physics engine, PhysX Vehicles.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[priyankanagaraj1494]

Hi @kraken24,

First of all, 0.9.4 has several issues with synchronous mode that have already being addressed in master, among them dead-locks when changing the map. Also take a look at #1803 as it may have some further fixes.

The thing you need to know is that every time the simulator does an update (advances the simulation one step), it sends a message to the clients that are connected. We also call this update a "tick".

• world.wait_for_tick() blocks until a new tick is received from the simulator. This function is not very reliable as you may skip ticks, and you can easily end up in a dead-lock if using synchronous mode.
• world.on_tick(callback) registers a callback (function or function object) to be called every time a new tick is received, the callback is executed asynchronously in the background.
• world.tick() tells the simulator to perform a tick (i.e. advance one step). This only has effect in synchronous mode.

So if you enable synchronous mode, the simulator is going to halt until you call world.tick().

Hi @nsubiron ,

What does the parameter seconds in the function world.wait_for_tick() mean?

[shuhangchen]

Hi @kraken24,

First of all, 0.9.4 has several issues with synchronous mode that have already being addressed in master, among them dead-locks when changing the map. Also take a look at #1803 as it may have some further fixes.

The thing you need to know is that every time the simulator does an update (advances the simulation one step), it sends a message to the clients that are connected. We also call this update a "tick".

• world.wait_for_tick() blocks until a new tick is received from the simulator. This function is not very reliable as you may skip ticks, and you can easily end up in a dead-lock if using synchronous mode.
• world.on_tick(callback) registers a callback (function or function object) to be called every time a new tick is received, the callback is executed asynchronously in the background.
• world.tick() tells the simulator to perform a tick (i.e. advance one step). This only has effect in synchronous mode.

So if you enable synchronous mode, the simulator is going to halt until you call world.tick().

Hi @nsubiron,

   I tried to control the car in synchronous mode, but the car was not moving.
   For example, in the example/synchronous_mode.py file, the current program "controls" the car by modifying the transform of the car directly. However, the car woud not react (move) if I use something like 

vehicle.apply_control(carla.VehicleControl(throttle=1.0, steer=0.0))

Can I ask if I did something wrong? or how to make the control work in this case?

Thanks!