Make On Demand Decision the default #3243
Conversation
| public void Awake() | ||
| { | ||
| m_Agent = gameObject.GetComponent<Agent>(); | ||
| Academy.Instance.AgentSetStatus += MakeRequests; |
There was a problem hiding this comment.
Should we rename Academy.AgentSetStatus too? Not really sure what makes sense for it.
There was a problem hiding this comment.
We could also remove it and use fixed update in the decision requester. It will change the logic slightly but would simplify greatly
There was a problem hiding this comment.
I think I like it better coming from the Academy, especially since users might not be stepping the Academy each FixedUpdate.
There was a problem hiding this comment.
Not sure about that, for Physics based environments, it is better to be synced with the Physics.
There was a problem hiding this comment.
In general that's true. But we shouldn't force that if the users knows (or thinks they know) better than us; that's why there's an option to disable automatic stepping on the Academy.
Maybe we should just clarify in the tooltips what "step" means here - "The agent will automatically request a decision every X Academy steps." and "Whether or not AgentAction will be called on Academy steps..."
There was a problem hiding this comment.
I agree with @chriselion here. We should drive the ticking through our systems, rather than have users subscribe to random ticking loops.
| } | ||
| void MakeRequests(int count) | ||
| { | ||
| if (count % DecisionPeriod == 0) |
There was a problem hiding this comment.
Either add a Range or Min attribute to DecisionPeriod, or take Mathf.max(DecisionPeriod, 1) (or both)
There was a problem hiding this comment.
I know it's "new" functionality, but I really think the ability to stagger which agents request each step would be really useful for developers with lots of agents. Otherwise you get large spikes every DecisionPeriod steps.
I think either something like
public int offset = 0
...
if ((count + offset) % DecisionPeriod == 0)or
public bool offsetStep = True;
...
var offset = offsetStep ? m_Agent.GetId() : 0; // or something similar
if ((count + offset) % DecisionPeriod == 0)The latter is probably more useful with prefabs since it doesn't require setting a different value on each.
There was a problem hiding this comment.
I know it's "new" functionality, but I really think the ability to stagger which agents request each step would be really useful for developers with lots of agents. Otherwise you get large spikes every DecisionPeriod steps.
This is a good candidate for thinking of jobification of ML-Agents. Not for 1.0, but later.
No description provided.