PTB: Robust physical backdoor attacks against deep neural networks in real world

•Compared to the backdoor attacks in digital domain, launching the backdoor attacks in real physical world are more challenging and more difficult.•We propose a robust physical backdoor attack with real physical triggers working under complex physical conditions.•The proposed method successfully addresses the influences of various 3D physical constraints.•Compared with normal backdoor attacks, the proposed attack method can significantly improve the attack performance in real physical world.•Three defenses which are effective against digital backdoor attacks fail to detect the proposed robust physical backdoor attack.