Topics of Interest
Topics of interest to this workshop include, but are not necessarily limited to:
- Mission/Motion Planning / Hierarchical/ Sampling-Based Planning
- Knowledge-Based Architecture / World Model and Awareness
- Deep Learning
- Self-Confidence / Assurance
- Adaptive / Collaborative Control
- Multi-Resolution Theory
In particular, the workshop and discussions will focus on following interrelated topics:
- Adaptive Motion Planning and Safety Guarantees : A planning system adaptively selects parameters or uses learnt models and heuristics to solve high dimensional problems in real- time. This topic features the interactions between machine learning and motion planning modules lead to complex problems that need to be addressed. One such problem is the train- test mismatch that occurs as the decisions made by the planning module affect the data distribution observed by the learning module. Another such problem arises when a learning module, which minimizes empirical risk, makes a single misclassification that can result in the planner getting `stuck' perpetually in a failure state. A key impediment to closing the loop on complex systems is the lack of a safety guarantee. The problem is exacerbated by the presence of uncertainty in the output of each system. Recent developments in formal methods and Bayesian optimization have furthered the level of complexity that can be verified.
- Deep Reinforcement Learning for Robotics : Deep reinforcement learning techniques have shown impressive results in the area of learning policies for complex systems in presence of uncertainty. This topic explores the recent advances in policy gradient methods and safe reinforcement learning. These have shown promise for enabling robots to learn through their own trial and error, resulting in augmented capabilities in manipulation and mobility. Furthermore, equipping robots with the ability to learn would by-pass the need for what otherwise often ends up being time-consuming task specific programming.
- Machine Self-Confidence : There is growing recognition for the need to provide users with better insight into the 'competency boundaries' and complex decision making processes of autonomous systems. This topic explores the possibilities for augmenting user-machine interaction with communication of an autonomous agent's "sense of confidence", as well as human-understandable explanations of the decision made under uncertainty. Self-confidence reporting goes above and beyond merely assessing whether assigned tasks can be successfully accomplished. Rather, self-confidence generally sumarizes an agent's holistic assessment of robustness regarding its ability to achieve assigned goals.
- Knowledge-Based Architecture and Collaborative Control : Complex systems are tasked with intricate missions that involve mission/motion planning and collaborative swarm control in large, obstacle-rich environments with competing objectives and multiple logical/spatial/temporal constraints. Knowledge-based architectures, hierarchical planning techniques, and Markov-chain enabled decentralized swarm control have demonstrated tractable methods for planning and execution of complex missions in dynamic and uncertain environments. This topic explores the methodologies for enabling context agnosticism, scalable contingency management, probability density distribution control of a swarm through velocity fields, and multi-resolution theory to construct abstractions of the environment of different granularity.
We hope that this workshop will serve as a medium to reach consensus on open problems that arise from integration and upon closing the loop on complex intelligent systems.