While reinforcement learning (RL) has become a more popular approach for robotics, designing sufficiently informative reward functions for complex tasks has proven to be extremely difficult due their inability to capture human intent and policy exploitation. Preference based RL algorithms seek to overcome these challenges by directly learning reward functions from human feedback. Unfortunately, prior work either requires an unreasonable number of queries implausible for any human to answer or overly restricts the class of reward functions to guarantee the elicitation of the most informative queries, resulting in models that are insufficiently expressive for realistic robotics tasks. Contrary to most works that focus on query selection to \emph{minimize} the amount of data required for learning reward functions, we take an opposite approach: \emph{expanding} the pool of available data by viewing human-in-the-loop RL through the more flexible lens of multi-task learning. Motivated by the success of meta-learning, we pre-train preference models on prior task data and quickly adapt them for new tasks using only a handful of queries. Empirically, we reduce the amount of online feedback needed to train manipulation policies in Meta-World by 20$\times$, and demonstrate the effectiveness of our method on a real Franka Panda Robot. Moreover, this reduction in query-complexity allows us to train robot policies from actual human users. Videos of our results and code can be found at https://sites.google.com/view/few-shot-preference-rl/home.

Valuing mineral assets is a challenging task that is highly dependent on the supply (geological) uncertainty surrounding resources and reserves, and the uncertainty of demand (commodity prices). In this work, a graph-based reasoning, modeling and solution approach is proposed to jointly address mineral asset valuation and mine plan scheduling and optimization under supply and demand uncertainty in the "mining complex" framework. Three graph-based solutions are proposed: (i) a neural branching policy that learns a block-sampling ore body representation, (ii) a guiding policy that learns to explore a heuristic selection tree, (iii) a hyper-heuristic that manages the value/supply chain optimization and dynamics modeled as a graph structure. Results on two large-scale industrial mining complexes show a reduction of up to three orders of magnitude in primal suboptimality, execution time, and number of iterations, and an increase of up to 40% in the mineral asset value.

Protein structure prediction is a fundamental problem in computational molecular biology. Classical algorithms such as ab-initio or threading as well as many learning methods have been proposed to solve this challenging problem. However, most reinforcement learning methods tend to model the state-action pairs as discrete objects. In this paper, we develop a reinforcement learning (RL) framework in a continuous setting and based on a stochastic parametrized Hamiltonian version of the Pontryagin maximum principle (PMP) to solve the side-chain packing and protein-folding problem. For special cases our formulation can be reduced to previous work where the optimal folding trajectories are trained using an explicit use of Langevin dynamics. Optimal continuous stochastic Hamiltonian dynamics folding pathways can be derived with use of different models of molecular energetics and force fields. In our RL implementation we adopt a soft actor-critic methodology however we can replace this other RL training based on A2C, A3C or PPO.

The Sinkhorn algorithm (arXiv:1306.0895) is the state-of-the-art to compute approximations of optimal transport distances between discrete probability distributions, making use of an entropically regularized formulation of the problem. The algorithm is guaranteed to converge, no matter its initialization. This lead to little attention being paid to initializing it, and simple starting vectors like the n-dimensional one-vector are common choices. We train a neural network to compute initializations for the algorithm, which significantly outperform standard initializations. The network predicts a potential of the optimal transport dual problem, where training is conducted in an adversarial fashion using a second, generating network. The network is universal in the sense that it is able to generalize to any pair of distributions of fixed dimension after training, and we prove that the generating network is universal in the sense that it is capable of producing any pair of distributions during training. Furthermore, we show that for certain applications the network can be used independently.

In recent years, unmanned aerial vehicle (UAV) related technology has expanded knowledge in the area, bringing to light new problems and challenges that require solutions. Furthermore, because the technology allows processes usually carried out by people to be automated, it is in great demand in industrial sectors. The automation of these vehicles has been addressed in the literature, applying different machine learning strategies. Reinforcement learning (RL) is an automation framework that is frequently used to train autonomous agents. RL is a machine learning paradigm wherein an agent interacts with an environment to solve a given task. However, learning autonomously can be time consuming, computationally expensive, and may not be practical in highly-complex scenarios. Interactive reinforcement learning allows an external trainer to provide advice to an agent while it is learning a task. In this study, we set out to teach an RL agent to control a drone using reward-shaping and policy-shaping techniques simultaneously. Two simulated scenarios were proposed for the training; one without obstacles and one with obstacles. We also studied the influence of each technique. The results show that an agent trained simultaneously with both techniques obtains a lower reward than an agent trained using only a policy-based approach. Nevertheless, the agent achieves lower execution times and less dispersion during training.

We develop a simple framework to learn bio-inspired foraging policies using human data. We conduct an experiment where humans are virtually immersed in an open field foraging environment and are trained to collect the highest amount of rewards. A Markov Decision Process (MDP) framework is introduced to model the human decision dynamics. Then, Imitation Learning (IL) based on maximum likelihood estimation is used to train Neural Networks (NN) that map human decisions to observed states. The results show that passive imitation substantially underperforms humans. We further refine the human-inspired policies via Reinforcement Learning (RL) using the on-policy Proximal Policy Optimization (PPO) algorithm which shows better stability than other algorithms and can steadily improve the policies pretrained with IL. We show that the combination of IL and RL can match human results and that good performance strongly depends on combining the allocentric information with an egocentric representation of the environment.

Many real-world continuous control problems are in the dilemma of weighing the pros and cons, multi-objective reinforcement learning (MORL) serves as a generic framework of learning control policies for different preferences over objectives. However, the existing MORL methods either rely on multiple passes of explicit search for finding the Pareto front and therefore are not sample-efficient, or utilizes a shared policy network for coarse knowledge sharing among policies. To boost the sample efficiency of MORL, we propose Q-Pensieve, a policy improvement scheme that stores a collection of Q-snapshots to jointly determine the policy update direction and thereby enables data sharing at the policy level. We show that Q-Pensieve can be naturally integrated with soft policy iteration with convergence guarantee. To substantiate this concept, we propose the technique of Q replay buffer, which stores the learned Q-networks from the past iterations, and arrive at a practical actor-critic implementation. Through extensive experiments and an ablation study, we demonstrate that with much fewer samples, the proposed algorithm can outperform the benchmark MORL methods on a variety of MORL benchmark tasks.

In this paper, we design a resource management scheme to support stateful applications, which will be prevalent in 6G networks. Different from stateless applications, stateful applications require context data while executing computing tasks from user terminals (UTs). Using a multi-tier computing paradigm with servers deployed at the core network, gateways, and base stations to support stateful applications, we aim to optimize long-term resource reservation by jointly minimizing the usage of computing, storage, and communication resources and the cost from reconfiguring resource reservation. The coupling among different resources and the impact of UT mobility create challenges in resource management. To address the challenges, we develop digital twin (DT) empowered network planning with two elements, i.e., multiresource reservation and resource reservation reconfiguration. First, DTs are designed for collecting UT status data, based on which UTs are grouped according to their mobility patterns. Second, an algorithm is proposed to customize resource reservation for different groups to satisfy their different resource demands. Last, a Meta-learning-based approach is developed to reconfigure resource reservation for balancing the network resource usage and the reconfiguration cost. Simulation results demonstrate that the proposed DT-empowered network planning outperforms benchmark frameworks by using less resources and incurring lower reconfiguration costs. C. Zhou, X. Shen, and W. Zhuang are with the Department of Electrical and Computer Engineering, University of Waterloo, Waterloo, ON, N2L 3G1, Canada (e-mail: c89zhou@uwaterloo.ca; J. Gao is with the School of Information Technology, Carleton University, Ottawa, ON, K1S 5B6, Canada (email: jie.gao6@carleton.ca).

We present a novel outdoor navigation algorithm to generate stable and efficient actions to navigate a robot to reach a goal. We use a multi-stage training pipeline and show that our approach produces policies that result in stable and reliable robot navigation on complex terrains. Based on the Proximal Policy Optimization (PPO) algorithm, we developed a novel method to achieve multiple capabilities for outdoor navigation tasks, namely alleviating the robot's drifting, keeping the robot stable on bumpy terrains, avoiding climbing on hills with steep elevation changes, and avoiding collisions. Our training process mitigates the reality (sim-to-real) gap by introducing generalized environmental and robotic parameters and training with rich features of Lidar perception in a high-fidelity Unity simulator. We evaluate our method in both simulation and real world environments using Clearpath Husky and Jackal robots. Further, we compare our method against the state-of-the-art approaches and observe that, in the real world it improves stability by at least 30.7% on uneven terrains, reduces drifting by 8.08% and decreases the elevation changes by 14.75%.

Learning policies that effectively utilize language instructions in complex, multi-task environments is an important problem in sequential decision-making. While it is possible to condition on the entire language instruction directly, such an approach could suffer from generalization issues. In our work, we propose \emph{Learning Interpretable Skill Abstractions (LISA)}, a hierarchical imitation learning framework that can learn diverse, interpretable primitive behaviors or skills from language-conditioned demonstrations to better generalize to unseen instructions. LISA uses vector quantization to learn discrete skill codes that are highly correlated with language instructions and the behavior of the learned policy. In navigation and robotic manipulation environments, LISA outperforms a strong non-hierarchical Decision Transformer baseline in the low data regime and is able to compose learned skills to solve tasks containing unseen long-range instructions. Our method demonstrates a more natural way to condition on language in sequential decision-making problems and achieve interpretable and controllable behavior with the learned skills.