Broad Problem: End-to-end real-time motion planning and control of autonomous agents via generative AI.
Key Challenges: Motion trajectories that satisfy fundamental constraints and guarantees (e.g., safety, stability, collision-avoidance) are difficult to generate with AI methods, especially while making it computationally efficient enough to run in real-time.
Research Goal: Develop planning algorithms for autonomous agents in various complex scenarios while maintaining fundamental constraints and guarantees (e.g., safety, stability, collision-avoidance), by combining both generative AI and traditional planning methods.
Application Areas: Robotics; Autonomous driving; Wheeled mobile robots; Robot Arm Manipulation.

Broad Problem: Large-scale, distributed or decentralized control and state-estimation for networks experiencing stochastic effects, such as external environment uncertainties and unmodeled dynamics.
Key Challenges: Scalable implementations of stochastic control and state-estimation algorithms introduce additional uncertainties (e.g., dynamic graph topology, delayed information transmission among agents) and additional constraints (e.g., limited communication bandwidth, global coordination among agents).
Research Goal: Develop distributed/decentralized stochastic control and state-estimation algorithms for large-scale systems that are robust, adaptive, and scalable (i.e., localizable) to high-dimensional network uncertainties.
Application Areas: Multi-Agent Systems; Large-Scale Networks; Fault-Tolerant Systems; Power Grid; Vehicle Traffic Grid; Sensor Networks.

Broad Problem: Real-time situational awareness via state-estimation for large-scale environments via wireless sensor networks.
Key Challenges: In conventional distributed data-gathering networks, the external sensors transmit all their measurements to the decision-making central agent, thereby increasing traffic congestion and unnecessarily redundant communications.
Research Goal: Develop a modular distributed data-gathering architecture with feedback communication from the central decision-maker to the external sensors, with theoretical and empirical guarantees on attaining a more optimal balance of efficiency and accuracy compared to architectures without feedback.
Application Areas: Distributed Data-Gathering; Multi-Agent Systems; Multi-Target Tracking; Wireless Sensor Networks; Robotic Sensor Networks.

Broad Problem: AGI for autonomous decision-making (e.g., embodied AI, continuous control) which adapts online to sequential tasks over time without catastrophically forgetting old tasks.
Key Challenges: Despite most real-world tasks containing repetitive structures or similar task patterns, many current AGI models retrain the existing model, making it sample- and memory-inefficient.
Research Goal: Develop theoretical and empirical characterizations of “task similarity” which can be incorporated into the learning process of AGI agents for autonomous decision-making, enabling them to achieve a more optimal balance between stability and plasticity while maintaining better memory- and sample-efficiency.
Application Areas: Embodied AI; Continual Learning; Continuous Control; Robotics; Robot Arm Manipulation.

Broad Problem: Preprocessing of raw training/fine-tuning data via representation learning and feature encoding, to reduce the dimensionality of the agent’s task.
Key Challenges: Many encoders and representation learning methods are designed to leverage exact patterns (e.g., group equivariance, invariance), which can yield a rigid representation of the data, thereby causing poor generalization and a lack of robustness as the agent performs its task.
Research Goal: Develop feature encoders and representation learning methods that search for approximate patterns or patterns that are a mixture of exact patterns, thereby enhancing the performance of AGI agents for autonomous decision-making while maintaining memory- and sample-efficiency.
Application Areas: Classification; Out-of-Distribution (OOD) Detection; Continual Learning; Continuous Control.