Eric Hsiung

Reward machines have shown great promise at capturing non-Markovian reward functions for learning tasks that involve complex action sequencing. However, no algorithm currently exists for learning reward machines with realistic weak feedback in the form of preferences. We contribute REMAP, a novel algorithm for learning reward machines from preferences, with correctness and termination guarantees. REMAP introduces preference queries in place of membership queries in the L* algorithm, and leverages a symbolic observation table along with unification and constraint solving to narrow the hypothesis reward machine search space. In addition to the proofs of correctness and termination for REMAP, we present empirical evidence measuring correctness: how frequently the resulting reward machine is isomorphic under a consistent yet inexact teacher, and the regret between the ground truth and learned reward machines.

As robots become more prevalent in society, they will need to learn to act appropriately under diverse human teaching styles. We present a human-centered approach for teaching robots reward functions by using a mixture of teaching strategies when communicating action appropriateness and goal success. Our method incorporates two teaching strategies for learning: explicit action instruction and evaluative, scalar-based feedback. We demonstrate that a robot instantiating our method can learn from humans who use both kinds of strategies to train the robot in a complex navigation task that includes norm-like constraints.

People are increasingly interacting with artificial agents in social settings, and as these agents become more sophisticated, people will have to teach them social norms. Two prominent teaching methods include instructing the learner how to act, and giving evaluative feedback on the learner’s actions. Our empirical findings indicate that people naturally adopt both methods when teaching norms to a simulated robot, and they use the methods selectively as a function of the robot’s perceived expertise and learning progress. In our algorithmic work, we conceptualize a set of context-specific norms as a reward function and integrate learning from the two teaching methods under a single likelihood-based algorithm, which estimates a reward function that induces policies maximally likely to satisfy the teacher’s intended norms. We compare robot learning under various teacher models and demonstrate that a robot responsive to both teaching methods can learn to reach its goal and minimize norm violations in a navigation task for a grid world. We improve the robot’s learning speed and performance by enabling teachers to give feedback at an abstract level (which rooms are acceptable to navigate) rather than at a low level (how to navigate any particular room).

Deep reinforcement learning has been extensively studied, resulting in several extensions to DQN that improve its perfor- mance, such as replay buffer sampling strategies, distributional value representations, and double/dueling networks. Previ- ous works have examined these extensions in the context of either discrete action spaces or in conjunction with actor-critic learning algorithms, but there has been no investigation of combining them for deep value-based continuous control. We adapted the methods discussed in Rainbow DQN to RBF-DQN, a deep valued-based method for continuous control, showing improvements in baseline performance and sample efficiency. Rainbow RBF-DQN is able to outperform vanilla RBF-DQN on the most challenging tasks even outperforming state of the art policy gradient methods like SAC.

Recent work on using natural language to specify commands to robots has grounded that language to LTL. However, mapping natural language task specifications to LTL task specifications using language models require probability distributions over finite vocabulary. Existing state-of-the-art methods have extended this finite vocabulary to include unseen terms from the input sequence to improve output generalization. However, novel out-of-vocabulary atomic propositions cannot be generated using these methods. To overcome this, we introduce an intermediate contextual query representation which can be learned from single positive task specification examples, associating a contextual query with an LTL template. We demonstrate that this intermediate representation allows for generalization over unseen object references, assuming accurate groundings are available. We compare our method of mapping natural language task specifications to intermediate contextual queries against state-of-the-art CopyNet models capable of translating natural language to LTL, by evaluating whether correct LTL for manipulation and navigation task specifications can be output, and show that our method outperforms the CopyNet model on unseen object references. We demonstrate that the grounded LTL our method outputs can be used for planning in a simulated OO-MDP environment. Finally, we discuss some common failure modes encountered when translating natural language task specifications to grounded LTL.

Learning continuous control in high-dimensional sparse reward settings, such as robotic manipulation, is a challenging problem due to the number of samples often required to obtain accurate optimal value and policy estimates. While many deep reinforcement learning methods have aimed at improving sample efficiency through replay or improved exploration techniques, state of the art actor-critic and policy gradient methods still suffer from the hard exploration problem in sparse reward settings. Motivated by recent successes of value-based methods for approximating state-action values, like RBF-DQN, we explore the potential of value-based reinforcement learning for learning continuous robotic manipulation tasks in multi-task sparse reward settings. On robotic manipulation tasks, we empirically show RBF-DQN converges faster than current state of the art algorithms such as TD3, SAC, and PPO. We also perform ablation studies with RBF-DQN and have shown that some enhancement techniques for vanilla Deep Q learning such as Hindsight Experience Replay (HER) and Prioritized Experience Replay (PER) can also be applied to RBF-DQN. Our experimental analysis suggests that value-based approaches may be more sensitive to data augmentation and replay buffer sample techniques than policy-gradient methods, and that the benefits of these methods for robot manipulation are heavily dependent on the transition dynamics of generated subgoal states.