Detector Threshold With Multiplicative Noise: The histograms of the q_k using the MLQG and LQG estimators with σ²_{a,1} = σ²_{c,1} = 0.06 are shown in red and cyan colors respectively. The moment based polynomial g(q) shown in blue and red curves bound their indicator functions in shaded blue and red respectively. Though both MLQG and LQG achieve mean-square compensation, the MLQG results in a tighter threshold than the LQG.
Robust Data-Driven Output Feedback Control via Bootstrapped Multiplicative Noise
Difference between CE and RMN on infinite-horizon performance J_T/J* and closed-loop spectral radius ρ(Φ_T) metrics vs time. Differences between metrics using CE and RMN were computed for each Monte Carlo trial individually; statistics of the resulting empirical distribution are shown.
Risk Bounded Nonlinear Robot Motion Planning With Integrated Perception & Control
Our proposed approach is demonstrated using an urban driving simulator [17]. An NRB-RRT* tree (in orange color) as a result of 200 iterations and the resulting reference path (in yellow color) from the start to the goal region are shown here. The NRB-RRT* algorithm employed distributionally robust chance constraints with α = 0.05.
Policy Iteration for Multiplicative Noise Output Feedback Control
Tree with 865 nodes grown by our proposed RANS-RRT* algorithm. Trajectory segments are shown as blue curves. The distributionally robust collision checks are represented by orange circles. The tree root is indicated by the white triangle and the goal region is the green rectangle with a dashed edge. Obstacles, including environment bounds are the black rectangles.
Stochastic Stability via Robustness of Linear Systems
Certified ms-stability levels βi found by Algorithm 1 (thm1) and the SDP (16) (sdp4). As a varies linearly from a = 0.001 to a = 1 paths are swept out from near the origin to larger positive values.
Learning Robust Control for LQR systems with Multiplicative Noise via Policy Gradient
Relative cost error (C(K)−C(K*))/C(K*) vs. iteration during policy gradient methods on a 4-state, 4-input lossy diffusion network with multiplicative noise using a) backtracking line search and b) constant step sizes.
Robust Control Design for Linear Systems via Multiplicative Noise
