BREAKING
Stanford open-sources GRaD-Nav
3DGS meets differentiable RL
3DGS
High-fidelity scenes
Differentiable representation
DiffRL
Gradients flow through sim
Faster policy learning
0%
more efficient
0%
of training time
GRaD-Nav++ success rates
Sim trained83
Sim unseen75
HW trained67
HW unseen50
VLA pipeline runs fully onboard
1Vision + language in
2MoE routing
3Actions out
4Onboard flight
A reproducible drone baseline
AI NEWS BLITZ
A Stanford team has released open-source code for vision-based drone navigation using 3D Gaussian Splatting.