Hi @RenzKa ,
Congrats for the outstanding work! I have two quick inquiries.
-
What is the time-step duration that you have considered in PlanT's design?
-
I understand that PlanT's planned trajectory is represented by a sequence of 2D waypoints (in BEV space) covering the future W=4 time-steps. Have you ever tested different lengths of the prediction window W? Would it be possible to re-train PlanT with a larger value of W? I guess that the value of W is constrained by the characteristics of the expert policy and that changing the length of the planned trajectory would require modifying the expert policy and generating a new dataset, but I am looking forward to your feedback.
Thank you in advance.
Best,
Luca
Hi @RenzKa ,
Congrats for the outstanding work! I have two quick inquiries.
What is the time-step duration that you have considered in PlanT's design?
I understand that PlanT's planned trajectory is represented by a sequence of 2D waypoints (in BEV space) covering the future W=4 time-steps. Have you ever tested different lengths of the prediction window W? Would it be possible to re-train PlanT with a larger value of W? I guess that the value of W is constrained by the characteristics of the expert policy and that changing the length of the planned trajectory would require modifying the expert policy and generating a new dataset, but I am looking forward to your feedback.
Thank you in advance.
Best,
Luca