Here's a video of the previously discussed obstacle avoidance scheme.
The obstacle avoidance method presented in this video uses time-of-impact sampling around the agent to adjust the agent's desired movement velocity (black arrow) to a direction of less potential collision (yellow arrow).
The figure around the agent is displaying time of collision in in each sampling direction (64 directions in the video). TOI up to a max time horizon (yellow circle) is considered while avoiding the other agents. If the TOI is less than the min time horizon (red dashed circle, time it takes to stop the agent), the steering velocity is scaled down.
Cool stuff
ReplyDeleteWill the approach be the same for 3D objects?
ReplyDeleteIn theory it should work for 3D stuff too. I'm not sure how I would change the sampling pattern, though.
ReplyDeleteI think having better sampling pattern would be a really good idea in general. For example to focus the samples on the last best velocity. And have less samples on the peripheral area.
This is rather splendid!
ReplyDelete