AI Researcher
Publications
Arthur Wandzel JAMM AI Framework (Nomi) Industry White Paper (2021).
At the heart of JAMM is the AI generating data insights for drivers and insurers (code named Nomi). Interestingly, the technology of Nomi follows two emerging trends in the early 2020s: 1). the cheapening of IOT applications and edge-based computing 2). the spread of automated, data-driven decision-making to new verticals (e.g. insurance). The result is a new insurtech product, JAMM, that yields insights underpinned by AI-derived, driver-road-context, redeemable for insurance savings. In this paper, we go over the essential JAMM AI architecture.
Keywords: AI architectures, attention, cloud computing, gaze estimation, human-behavioral-understanding, industry application, object tracking and recognition
Arthur Wandzel, Panpan Cai, and David Hsu, Perception for Planning: Integrating Attention into Planning for Deep Robot
Navigation, Technical Report (2021).
In merging on a highway or rounding a traffic-circle, humans identify other vehicles based on selective attention—filtering out the irrelevant vehicles to spend precious cognitive resources on the relevant. In this work, we introduce a principled approach for learning attention weights for motion planning under uncertainty. We train a custom graph neural network (GNN) model on labels derived from a model-based planner (Gamma). This approach allows us to learn a context-dependent probability distribution over vehicles to filter monte-carlo simulations for forward planning. In doing so, we demonstrate compelling results: reduced collision rates and faster travel times in simulation while under real-time decision-making constraints (< 50 milliseconds).
Keywords: autonomous vehicles, attention, deep learning, graph neural networks, planning and learning, robotics
Arthur Wandzel, Seungchan Kim, Stefanie Tellex, and Yoonseon Oh, Sample Bounds for Robust Multi-Object POMDP Planning via Rademacher Complexity, Preprint (2020).
Object-based reasoning in real-world environments imposes a challenge: as the number of considered objects scale, planning becomes increasingly computationally intractable. In this paper, we derive general upper bounds on the number of samples for Q-value estimation in the context of object-based, online POMDP planning. Our bounds feature a novel application of Rademacher complexity for POMDPs, which comprises of two terms: a regularization term that penalizes complex POMDP models and a counting term that scales with the size of the POMDP problem. We compare bounds as we vary model factorization in terms of objects. We conclude by empirically validating our theoretical findings by demonstrating the advantage of belief factorization for supporting sample-efficient multi-object POMDP planning on a number of domains.
Keywords: computational learning theory, POMDPs, planning, object-based reasoning, Rademacher complexity
Arthur Wandzel, Yoonseon Oh, Michael Fishman, Nishanth Kumar, Lawson L.S. Wong, and Stefanie Tellex, Multi-Object Search using Object-Oriented POMDPs, IEEE International Conference on Robotics and Automation (ICRA 2019).
A core capability of robots is to reason about multiple objects under uncertainty. Partially Observable Markov Decision Processes (POMDPs) provide a means of reasoning under uncertainty for sequential decision making, but are computationally intractable in large domains. In this paper, we propose Object-Oriented POMDPs (OO-POMDPs), which represent the state and observation spaces in terms of classes and objects. The structure afforded by OO-POMDPs support a factorization of the agent’s belief into independent object distributions, which enables the size of the belief to scale linearly versus exponentially in the number of objects. We formulate a novel Multi-Object Search (MOS) task as an OO-POMDP for mobile robotics domains in which the agent must find the locations of multiple objects. Our solution exploits the structure of OO-POMDPs by featuring human language to selectively update the belief at task onset. Using this structure, we develop a new algorithm for efficiently solving OO-POMDPs: Object- Oriented Partially Observable Monte-Carlo Planning (OO- POMCP). We show that OO-POMCP with grounded language commands is sufficient for solving challenging MOS tasks both in simulation and on a physical mobile robot.
Keywords: collaborative robotics, grounded language commands, object-based search, POMDPs, planning
Steven J. Jones, Arthur Wandzel, and John E. Laird, Efficient Computation of Spreading Activation Using Lazy Evaluation, International Conference on Cognitive Modeling (ICCM 2016).
Spreading activation is an important component of many computational models of declarative long-term memory retrieval but it can be computationally expensive. The computational overhead has led to severe restrictions on its use, especially in real-time cognitive models. In this paper we describe a series of successively more efficient algorithms for spreading activation. The final model uses lazy evaluation to avoid much of the computation normally associated with spreading activation. We evaluate its efficiency on a commonly-used word sense disambiguation task where it is significantly faster than a naive model, achieving an average time of 0.43ms per query for a spread to 300 nodes.
Keywords: cognitive modeling, probabalistic methods, memory retrieval, SOAR