Community Profile

photo

Emmanouil Tzorakoleftherakis

MathWorks

Last seen: 4 giorni fa Attivo dal 2018

Statistiche

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

Visualizza badge

Content Feed

Visto da

Risposto
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...

8 giorni fa | 0

| accettato

Risposto
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.

29 giorni fa | 0

| accettato

Risposto
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...

circa un mese fa | 0

Risposto
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...

circa un mese fa | 0

Risposto
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 ...

circa un mese fa | 0

Risposto
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...

circa un mese fa | 1

| accettato

Risposto
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 ...

circa un mese fa | 0

| accettato

Risposto
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...

circa 2 mesi fa | 0

| accettato

Risposto
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...

circa 2 mesi fa | 0

Risposto
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...

circa 2 mesi fa | 0

Risposto
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...

circa 2 mesi fa | 0

| accettato

Risposto
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...

circa 2 mesi fa | 0

| accettato

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

circa 2 mesi fa | 0

| accettato

Risposto
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...

circa 2 mesi fa | 0

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

2 mesi fa | 0

| accettato

Risposto
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...

2 mesi fa | 0

| accettato

Risposto
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.

2 mesi fa | 1

| accettato

Risposto
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...

2 mesi fa | 1

Risposto
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...

2 mesi fa | 0

Risposto
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...

2 mesi fa | 1

| accettato

Risposto
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...

2 mesi fa | 0

| accettato

Risposto
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...

2 mesi fa | 0

| accettato

Risposto
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...

2 mesi fa | 0

| accettato

Risposto
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 ...

3 mesi fa | 0

Risposto
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...

3 mesi fa | 0

Risposto
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?

3 mesi fa | 0

| accettato

Risposto
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...

3 mesi fa | 0

Risposto
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...

3 mesi fa | 0

Risposto
PPO and LSTM agent creation
Hi, With lstm policies, BOTH the actor and the critic should have lstm layers. That's why you are getting this error. LSTM po...

3 mesi fa | 0

Risposto
Matlab reinforcement learning stop working after a while
125 episodes are not that many. Is it always freezing on that episode? Given that you are not getting any errors/crashes, my hun...

3 mesi fa | 0

Carica altro