Finding the discrete to continuous action translation

[k4ntz]

Hi, I am trying to understand how the inputted action changes the state in Pong and Breakout.
I know that these games were usually controlled using the knob controller (cf here on the right of the image).

Such a controller directly provides the position of the paddle. As ALE uses by default discrete actions, these are translated to continuous ones in this environment. Thus, a state of the Paddle is maintained somewhere. ALE takes in the action, modifies the speed or the acceleration of the Paddle, and thus update the position of the paddle, sent to the game.

I am looking for the function that match this Paddle_state + action to the paddle position.

I have been looking into src/ale/environment/ale_state.cpp, but could not find the matching function.

Do you know where it is executed ?

[pseudo-rnd-thoughts]

Why isn't it updatePadlePositions -

Arcade-Learning-Environment/src/ale/environment/ale_state.cpp

Line 168 in 6b57d42

void ALEState::updatePaddlePositions(Event* event, int delta_left,

[k4ntz]

Yes, it is called in ALEState::applyActionPaddles.
In short, you have:

      delta_a = static_cast<int>(PADDLE_DELTA * fabs(paddle_a_strength)); 
      updatePaddlePositions(event, delta_a, delta_b);

And in u:

void ALEState::updatePaddlePositions(Event* event, int delta_left,
                                     int delta_right) {
    m_left_paddle += delta_left;
   ...

Thus, I started printing (with cout) the values of delta_left, but they are always constant in my tries (0 for NOOP) or constant for DOWN or UP actions.

I also looked into:

void ALEState::setPaddles(Event* event, int left, int right) {
  m_left_paddle = left;
  m_right_paddle = right;

  // Compute the "resistance" (this is for vestigal clarity)
  int left_resistance = calcPaddleResistance(m_left_paddle);
  int right_resistance = calcPaddleResistance(m_right_paddle);

  // Update the events with the new resistances
  event->set(Event::PaddleZeroResistance, left_resistance);
  event->set(Event::PaddleOneResistance, right_resistance);
}

But it seems that this resistance is then never used. As the delta I see applied are constant.

[pseudo-rnd-thoughts]

Shouldn't you care about m_left_paddle and m_right_paddle rather than the delta as this is the amount that the paddles move.