A dynamics-aware, look-ahead attack on the action space of deep RL agents consistently beats a myopic action-space attack at equal total budget and also reveals which actuators are most vulnerable.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Spatiotemporally Constrained Action Space Attacks on Deep Reinforcement Learning Agents
A dynamics-aware, look-ahead attack on the action space of deep RL agents consistently beats a myopic action-space attack at equal total budget and also reveals which actuators are most vulnerable.