photo

Emmanouil Tzorakoleftherakis

MathWorks

Last seen: 19 jours il y a |  Actif depuis 2018

Followers: 3   Following: 0

Statistiques

All
  • Thankful Level 3
  • 12 Month Streak
  • Personal Best Downloads Level 1
  • Pro
  • Knowledgeable Level 5
  • GitHub Submissions Level 1
  • First Submission
  • Revival Level 2
  • First Answer

Afficher les badges

Feeds

Afficher par

Réponse apportée
Reinforcement Learning on Simscape
One option is to look at introducing the delay on the observation, not the action. Please take a look at <https://www.mathworks....

24 jours il y a | 0

| A accepté

Réponse apportée
Does MPC without constraints also solves QP problem? How to check QP solver within matlab?
Hi, If you don't have constraints, the toolbox will use the analytical solution to the problem, which you can see here. This ha...

environ 2 mois il y a | 0

Réponse apportée
Reaching observation data and pass them to the learning process
In general, you cannot change the observation/action space definition once they are defined. That said, it seems to me that what...

4 mois il y a | 0

| A accepté

Réponse apportée
How to build a reinforcement learning environment for a DCDC converter?
I would look at this example which starts with a model that has a PID controller and shows how to replace it with an RL Agent.

5 mois il y a | 0

| A accepté

Réponse apportée
PPO minibatch size for parallel training with variable number of steps
No data will be discarded actually. As of R2023b, the 4 experiences that are left in your example form their own minibatch and a...

5 mois il y a | 0

Réponse apportée
Why is my DDPG agent converging to a state where it gets continuous penalization, while having a state it can go with 0 penalization?
My guess is that this happens due to the specifics of the problem. You want to build a controller that generates 'zeroes' when t...

5 mois il y a | 0

Réponse apportée
Reinforcement learning: Step function "AvoidObstaclesUsingReinforcementLearningForMobileRobotsExample" example
This example trains the agent against a Simulink environment, not a MATLAB one. The equivalent of the 'step' function is inside ...

5 mois il y a | 0

Réponse apportée
How can I deploy the trained DRL model in a microprocessor, such as DSP or STM32?
You can follow the steps here to generate code from the trained policy. We also have hardware support for STM32 processors, so i...

5 mois il y a | 1

| A accepté

Réponse apportée
Parallel Training of Multiple RL Agents in same environment
Parallel training is currently not supported for multi-agent reinforcement learning. One thing you could do is train the agents ...

5 mois il y a | 0

| A accepté

Réponse apportée
Augmenting MPC Block with Integral Action
Hello, Let me paste a couple of links here that show how we formulate the underlying QP problem in linear mpc in Model Predicti...

5 mois il y a | 0

| A accepté

Réponse apportée
Does my PI + MPC (feedforward controller) configuration make sense?
Looks like the whole point of using an MPC controller was to provide deltas on the PI output based on the output of the piezo ac...

5 mois il y a | 0

Réponse apportée
how to freeze and reset the weights to initial values of neural network.?
You can accomplish what you asked with something along the lines of: init_model = getModel(getCritic(agent)); new_model_layers...

5 mois il y a | 0

Réponse apportée
Cannot propagate non-bus signal to block because the block has a bus object specified.
Looking at the first screenshot, looks like the output of the grid world block is not a bus, but the observations in your RL Age...

5 mois il y a | 0

| A accepté

Réponse apportée
Constraint to state derivatives with NLMPC
See here for all available constraint options with nlmpc. If the state derivatives you need are part of the state vector, you ca...

6 mois il y a | 0

| A accepté

Réponse apportée
Cannot generate C code from MPC object
Please take a look at the example here that uses 'codegen' command.

6 mois il y a | 0

| A accepté

Réponse apportée
RL Agent learns a constant trajectory instead of actual trajectory
Thanks for adding all the details. The first thing I will say is that the average reward on the Episode Manager is moving in the...

6 mois il y a | 0

Réponse apportée
Tune PI Controller Using Reinforcement Learning
Do you maybe have linearize shadowed somewhere on your path? If not, a reproduction model would be good.

6 mois il y a | 0

| A accepté

Réponse apportée
Training Reinforcement Learning Agents --> Use ResetFcn to delay the agent's behaviour in the environment
You can place the RL Agent block inside a triggered subsystem and set the agent's sample time to -1 (see e.g. here). Then set th...

6 mois il y a | 0

| A accepté

Réponse apportée
How to specify the training algorithm of an agent - Reinforcement Learning
'train' takes an agent object as input, so yes the algorithm will be selected depending on the agent.

6 mois il y a | 1

| A accepté

Réponse apportée
DDPG training converges to the worst results obtained during exploration
I cannot see your training options, but what do you mean by "converges"? The training plot only shows about 1800 episodes. There...

6 mois il y a | 1

Réponse apportée
Not able to use multibel GPUs when training a DDPG agent
Can you share your agent options and the architecture of the actor and critic networks? As mentioned here, "Using GPUs is likely...

6 mois il y a | 0

Réponse apportée
Problem with RL agent block
You can use a delay block for the last observation and set the initial value of the delay in the block dialog. That should resol...

6 mois il y a | 1

| A accepté

Réponse apportée
decaying clip factor or entropy loss weight for PPO
These parameters are fixed and cannot be changed after training begins. One workaround would be to train the agent for a certain...

6 mois il y a | 0

| A accepté

Réponse apportée
How to solve the error "Error using sqpInterface Nonlinear constraint function is undefined at initial point. Fmincon cannot continue." Error occurred when calling NLP solver
The dynamics/state function are turned into constraints internally when creating a NLP for fmincon. You don't provide all the fu...

6 mois il y a | 0

| A accepté

Réponse apportée
How do I Tune Model Predictive Controller (MPC) in the Real Time?
There could be many reasons why you don't see the expected results. First thing I would check is whether the controller can actu...

6 mois il y a | 0

| A accepté

Réponse apportée
How Fast is Simulink Real Time? Is Simulink Real Time Faster than Rasberry Pi?
Hi, First of all, how fast (wall clock time) does the Simulink model run with your current MPC implementation? This would be a ...

6 mois il y a | 0

Réponse apportée
Design an actor critic network for non-image inputs
I may be missing something but why don't you frame your observations as a [4 1] vector? That way it would be consistent with how...

7 mois il y a | 0

Réponse apportée
Pausing reinforcement learning by forcing
The proper way to stop it would be through the Episode Manager (top right of the window). Does this not work for you?

7 mois il y a | 0

| A accepté

Réponse apportée
Why DQN training always fails to converge to the optimal value
What I am seeing here is that the average reward tends to converge to the Q0 profile which is the expected behavior of a converg...

7 mois il y a | 0

Réponse apportée
My reinforcement learning simulation runs for only 0 steps and 0 times in Simulink. I am not getting any error messages so I cannot pinpoint the issue, so I decided to ask.
It likely has to do with the priority of execution of the data store blocks. I would look more into it, but honestly I think you...

7 mois il y a | 0

Charger plus