efficient collide detection improvements and questions

Official Python bindings with a focus on reinforcement learning and robotics.
Post Reply
Posts: 1
Joined: Wed Mar 01, 2023 7:54 am

efficient collide detection improvements and questions

Post by sancelot88 »

I am working on an offline collision program. The meshes are rigid bodies of a complex machine moving (models are convex meshes converted from creo/proe cad systems).

I have positions of each mesh depending on the time. For optimisation reasons, I have positions only when something is moving at least 0.5mm. This implies non constant delta times.
eg , positions for t0=0ms , t1=4ms, t2=10ms,t3=50ms ......(this is not real , but to illustrate ....)

I started with a basic collide algorithm in which at each time frame I am computing and setting a new physics world with the whole scene to perform collisions.
It works, but I is not really efficient regarding performance.

I would like now migrate to a continuous detection collision mode, in which I will set positions (transformation matrix) of each mesh, without reconstructiong the whole physics world. Which orientations and modifications should be involved regarding my initial implementation ?

Here is basically the code I am using for each frame (with ammojs but does not sound being a pb).

Code: Select all

 this._broadphase = new this.ammoInjection.btDbvtBroadphase();
 this._collisionConfiguration = new this.ammoInjection.btDefaultCollisionConfiguration();
 this._dispatcher = new this.ammoInjection.btCollisionDispatcher(this._collisionConfiguration);

 this._physicsWorld = new this.ammoInjection.btCollisionWorld(
for each mesh:
    const shape = new this.ammoInjection.btConvexTriangleMeshShape(trimesh, true);
    const m = new this.ammoInjection.btCollisionObject();
    this._physicsWorld.addCollisionObject(m, mesh.collisionGroup, mesh.collisionMask);


// collect manifolds 
Post Reply