SOTA perception algorithms have error rates that are higher than what is acceptable for safe autonomy
SOTA lacks robustness to erroneous/outlier data found in practice
Many algorithms cannot be used in practice due to their high computational complexity and lack of scalability
Multi-way & multi-modal inference, addressing the shortcomings of a single sensor modality
Algorithmic frameworks for robust data association and outlier rejection (robust to >99% outliers)
Fast & certifiable perception algorithms
P. Lusk, K. Fathian, J. P. How, “MIXER: Multiattribute, Multiway Fusion of Uncertain Pairwise Affinities,” IEEE Robotics and Automation Letters, 2022.
Viewpoint variations (opposite views, air-ground)
Changing or dynamic environments
Off-nominal conditions (bad weather, sensing failures)
View-invariant semantic & topological representations
Integration of SLAM and modern outlier rejection techniques to create robustness to environment changes, sensing failures, noisy and out-of-distribution data, etc.
Limited communication budget
Lack of global knowledge (distributed computing)
Adversarial agents/data
IoRT on 4G/5G cellular networks
Mixed edge & cloud computing (“fog computing”)
Distributed learning, perception, planning, control
Adversary-resilient multi-agent optimization
Planned trajectory may block camera view
Planner does not consider perception uncertainty
Coupled perception-planning-control algorithms (e.g., perception-aware planning)
Active, multi-robot SLAM
Fast C++ implementations, ROS packages, & field robotic experiments