CLAE steers frozen multi-quadrotor policies at inference by training a sparse autoencoder on activations, probing for relevant latents, and learning an RL policy to apply state-dependent affine edits for velocity control, formation preservation, and camera avoidance.
Title resolution pending
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
fields
cs.RO 2years
2026 2verdicts
UNVERDICTED 2representative citing papers
Crazyflow is a fast, differentiable, GPU-accelerated drone simulator in JAX that enables large-scale parallel simulation and in-flight reinforcement learning for aerial robots.
citing papers explorer
-
Steering Multirobot Behavior via Closed-Loop Affine Activation Editing
CLAE steers frozen multi-quadrotor policies at inference by training a sparse autoencoder on activations, probing for relevant latents, and learning an RL policy to apply state-dependent affine edits for velocity control, formation preservation, and camera avoidance.
-
Crazyflow: An Accurate, GPU-Accelerated, Differentiable Drone Simulator in JAX
Crazyflow is a fast, differentiable, GPU-accelerated drone simulator in JAX that enables large-scale parallel simulation and in-flight reinforcement learning for aerial robots.