Object-oriented representations in reinforcement learning have shown promise in transfer learning, with previous research introducing a propositional objectoriented framework that has provably efficient learning bounds with respect to samplecomplexity.

Post rebuttal: I now understand the middle ground this paper is positioned, and the difference to propositional OO representations where you don't necessarily care which instance of an object type you're dealing with, which significantly reduces the dimensionality of learning transition dynamics. But this is still similar to other work on graph neural networks for model learning in fully relational representations, like Relation Networks by Santoro et al., and Interaction Networks by Battaglia et al. which in worst case learn T * n * (n-1) relations for n objects for T types of relations. However, this paper does do a nice job of formalizing from the OO-MDP and Propositional MDP setting as opposed to the two papers I mentioned which do not, and focus on the physical dynamics case. I am willing to increase my score based on this, but still do not think it is novel enough to be accepted. This is very similar to relational MDPs, but they learn transition dynamics in this relational attribute space rather than real state space.

We present ongoing research to develop novel option discovery methods for complex domains that are represented as Object-Oriented Markov Decision Processes (OO-MDPs) (Diuk, Cohen, and Littman, 2008). We describe Portable Multi-policy Option Discovery for Automated Learning (P-MODAL), an initial framework that extends Pickett and Barto's (2002) PolicyBlocks approach to OO-MDPs. We also discuss future work that will use additional representations and techniques to handle scalability and learning challenges.

We present an initial proposal for modular object-oriented MDPs, an extension of OO-MDPs that abstracts complex domains that are partially observable and stochastic with multiple goals. Modes reduce the curse of dimensionality by reducing the number of attributes, objects, and actions into only the features relevant for each goal. These modes may also be used as an abstracted domain to be transferred to other modes or to another domain.

Massive state spaces are ubiquitous throughout planning and reinforcement learning (RL) domains: agents involved in furniture assembly, cooking automation and backgammon must grapple with problem formalisms that are much too expansive to solve by conventional tabular approaches. However, modern tabular planning and RL techniques bypass this difficulty by using propositional functions to transfer knowledge across states — both within and across problem instances — to solve for near optimal behaviors in very large state spaces. Here we present a means by which useful propositional functions can be inferred from observations of transition dynamics. Our approach is based upon distilling salient relational values between pairs of objects. We then use these learned propositional functions to free the RL algorithm deterministic object-oriented RMAX (DOORMAX) of its dependence on expert-provided propositional functions. We also empirically demonstrate high correspondence between these learned propositional functions and expert-provided propositional functions. Our novel DOORMAX algorithm performs at a level near that of classic DOORMAX.

Performing experimental research on robotic platforms involves numerous practical complications, while studying collaborative interactions and efficiently collecting data from humans benefit from real time response. Roboticists can circumvent some complications by using simulators like Gazebo to test algorithms and building games like the Mars Escape game to collect data. Making use of existing resources for simulation and game creation requires the development of assets and algorithms along with the recruitment and training of users. We have created a Minecraft mod called BurlapCraft which enables the use of the reinforcement learning and planning library BURLAP to model and solve different tasks within Minecraft. BurlapCraft makes AI-HRI development easier in three core ways: the underlying Minecraft environment makes the construction of experiments simple for the developer and so allows the rapid prototyping of experimental setup; BURLAP contributes a wide variety of extensible algorithms for learning and planning, allowing easy iteration and development of task models and algorithms; and the familiarity and ubiquity of Minecraft trivializes the recruitment and training of users. To validate BurlapCraft as a platform for AI development, we demonstrate the execution of A*, BFS, RMax, language understanding, and learning language groundings from user demonstrations in five Minecraft "dungeons."

We present ongoing research to develop novel option discovery methods for complex domains that are represented as Object-Oriented Markov Decision Processes (OO-MDPs) (Diuk, Cohen, and Littman, 2008). We describe Portable Multi-policy Option Discovery for Automated Learning (P-MODAL), an initial framework that extends Pickett and Barto’s (2002) PolicyBlocks approach to OO-MDPs. We also discuss future work that will use additional representations and techniques to handle scalability and learning challenges.

Robotic agents often map perceptual input to simplified representations that do not reflect the complexity and richness of the world. This simplification is due in large part to the limitations of planning algorithms, which fail in large stochastic state spaces on account of the well-known "curse of dimensionality." Existing approaches to address this problem fail to prevent autonomous agents from considering many actions which would be obviously irrelevant to a human solving the same problem. We formalize the notion of affordances as knowledge added to an Markov Decision Process (MDP) that prunes actions in a state- and reward- general way. This pruning significantly reduces the number of state-action pairs the agent needs to evaluate in order to act near-optimally. We demonstrate our approach in the Minecraft domain as a model for robotic tasks, showing significant increase in speed and reduction in state-space exploration during planning. Further, we provide a learning framework that enables an agent to learn affordances through experience, opening the door for agents to learn to adapt and plan through new situations. We provide preliminary results indicating that the learning process effectively produces affordances that help solve an MDP faster, suggesting that affordances serve as an effective, transferable piece of knowledge for planning agents in large state spaces.