Attack detection and countermeasures for autonomous navigation
Autonomous navigation cyber-attack detection and/or avoidance techniques include visual and inertial odometry (VIO) algorithms to provide a root-of-trust during navigation, VIO algorithms that cross-validate navigation parameters using IMU and visual data, and hardware-dependent attack survival mechanisms that support autonomous systems during an attack.
1. A method for detecting a replay attack on an autonomous navigation system and surviving said replay attack, comprising:
using two or more stereo image sensors mounted in an autonomous vehicle to collect primary pose data for said autonomous vehicle, using inertial measurement units mounted in said autonomous vehicle to collect secondary pose date for said autonomous vehicle,
sending said primary data and said secondary data to a neural network,
using an open-loop shallow neural network-based nonlinear autoregressive exogenous model on said neural network to train drift estimation between said primary pose data and said secondary pose data; and
using a closed-loop model on said neural network for multi-step prediction of pose drift y(t) between said primary pose data and said secondary pose data based on the equation:
y
(
t
)
=
F
(
y
t
-
1
,
y
t
-
2
,
…
,
x
t
,
x
t
-
1
,
x
t
-
2
,
…
)
+
ϵ
;
(
1
)
wherein said neural network comprises a predefined threshold modeling error and a predefined temporal window for assessing anomalous events;
said method further comprising implementing attack survival steps when error E(t) exceeds said predefined threshold modeling error, said attack survival steps including termination of autonomous driving.
2. The method of claim 1 , further comprising monitoring trusted hardware mounted internally in said autonomous navigation system to detect data corruption by cross-validating GPS time signals with a free-running oscillator.