Reinforcement learning (RL) agents are particularly hard to train when rewards are sparse. One common solution is to use intrinsic rewards to encourage agents to explore their environment. However, recent intrinsic exploration methods often use state-based novelty measures which reward low-level exploration and may not scale to domains requiring more abstract skills. Instead, we explore natural language as a general medium for highlighting relevant abstractions in an environment. Unlike previous work, we evaluate whether language can improve over existing exploration methods by directly extending (and comparing to) competitive intrinsic exploration baselines: AMIGo (Campero et al., 2021) and NovelD (Zhang et al., 2021). These language-based variants outperform their non-linguistic forms by 47-85% across 13 challenging tasks from the MiniGrid and MiniHack environment suites.

In reinforcement learning, the performance of learning agents is highly sensitive to the choice of time discretization. Agents acting at high frequencies have the best control opportunities, along with some drawbacks, such as possible inefficient exploration and vanishing of the action advantages. The repetition of the actions, i.e., action persistence, comes into help, as it allows the agent to visit wider regions of the state space and improve the estimation of the action effects. In this work, we derive a novel All-Persistence Bellman Operator, which allows an effective use of both the low-persistence experience, by decomposition into sub-transition, and the high-persistence experience, thanks to the introduction of a suitable bootstrap procedure. In this way, we employ transitions collected at any time scale to update simultaneously the action values of the considered persistence set. We prove the contraction property of the All-Persistence Bellman Operator and, based on it, we extend classic Q-learning and DQN. After providing a study on the effects of persistence, we experimentally evaluate our approach in both tabular contexts and more challenging frameworks, including some Atari games.

Learning agile skills is one of the main challenges in robotics. To this end, reinforcement learning approaches have achieved impressive results. These methods require explicit task information in terms of a reward function or an expert that can be queried in simulation to provide a target control output, which limits their applicability. In this work, we propose a generative adversarial method for inferring reward functions from partial and potentially physically incompatible demonstrations for successful skill acquirement where reference or expert demonstrations are not easily accessible. Moreover, we show that by using a Wasserstein GAN formulation and transitions from demonstrations with rough and partial information as input, we are able to extract policies that are robust and capable of imitating demonstrated behaviors. Finally, the obtained skills such as a backflip are tested on an agile quadruped robot called Solo 8 and present faithful replication of hand-held human demonstrations.

Oversubscription is a common practice for improving cloud resource utilization. It allows the cloud service provider to sell more resources than the physical limit, assuming not all users would fully utilize the resources simultaneously. However, how to design an oversubscription policy that improves utilization while satisfying the some safety constraints remains an open problem. Existing methods and industrial practices are over-conservative, ignoring the coordination of diverse resource usage patterns and probabilistic constraints. To address these two limitations, this paper formulates the oversubscription for cloud as a chance-constrained optimization problem and propose an effective Chance Constrained Multi-Agent Reinforcement Learning (C2MARL) method to solve this problem. Specifically, C2MARL reduces the number of constraints by considering their upper bounds and leverages a multi-agent reinforcement learning paradigm to learn a safe and optimal coordination policy. We evaluate our C2MARL on an internal cloud platform and public cloud datasets. Experiments show that our C2MARL outperforms existing methods in improving utilization ($20\%\sim 86\%$) under different levels of safety constraints.

Realistic long-horizon tasks like image-goal navigation involve exploratory and exploitative phases. Assigned with an image of the goal, an embodied agent must explore to discover the goal, i.e., search efficiently using learned priors. Once the goal is discovered, the agent must accurately calibrate the last-mile of navigation to the goal. As with any robust system, switches between exploratory goal discovery and exploitative last-mile navigation enable better recovery from errors. Following these intuitive guide rails, we propose SLING to improve the performance of existing image-goal navigation systems. Entirely complementing prior methods, we focus on last-mile navigation and leverage the underlying geometric structure of the problem with neural descriptors. With simple but effective switches, we can easily connect SLING with heuristic, reinforcement learning, and neural modular policies. On a standardized image-goal navigation benchmark (Hahn et al. 2021), we improve performance across policies, scenes, and episode complexity, raising the state-of-the-art from 45% to 55% success rate. Beyond photorealistic simulation, we conduct real-robot experiments in three physical scenes and find these improvements to transfer well to real environments.

In the last few years, the research activity around reinforcement learning tasks formulated over environments with sparse rewards has been especially notable. Among the numerous approaches proposed to deal with these hard exploration problems, intrinsic motivation mechanisms are arguably among the most studied alternatives to date. Advances reported in this area over time have tackled the exploration issue by proposing new algorithmic ideas to generate alternative mechanisms to measure the novelty. However, most efforts in this direction have overlooked the influence of different design choices and parameter settings that have also been introduced to improve the effect of the generated intrinsic bonus, forgetting the application of those choices to other intrinsic motivation techniques that may also benefit of them. Furthermore, some of those intrinsic methods are applied with different base reinforcement algorithms (e.g. PPO, IMPALA) and neural network architectures, being hard to fairly compare the provided results and the actual progress provided by each solution. The goal of this work is to stress on this crucial matter in reinforcement learning over hard exploration environments, exposing the variability and susceptibility of avant-garde intrinsic motivation techniques to diverse design factors. Ultimately, our experiments herein reported underscore the importance of a careful selection of these design aspects coupled with the exploration requirements of the environment and the task in question under the same setup, so that fair comparisons can be guaranteed.

Backdoor attacks on reinforcement learning implant a backdoor in a victim agent's policy. Once the victim observes the trigger signal, it will switch to the abnormal mode and fail its task. Most of the attacks assume the adversary can arbitrarily modify the victim's observations, which may not be practical. One work proposes to let one adversary agent use its actions to affect its opponent in two-agent competitive games, so that the opponent quickly fails after observing certain trigger actions. However, in multiagent collaborative systems, agents may not always be able to observe others. When and how much the adversary agent can affect others are uncertain, and we want the adversary agent to trigger others for as few times as possible. To solve this problem, we first design a novel training framework to produce auxiliary rewards that measure the extent to which the other agents'observations being affected. Then we use the auxiliary rewards to train a trigger policy which enables the adversary agent to efficiently affect the others' observations. Given these affected observations, we further train the other agents to perform abnormally. Extensive experiments demonstrate that the proposed method enables the adversary agent to lure the others into the abnormal mode with only a few actions.

An important goal in artificial intelligence is to create agents that can both interact naturally with humans and learn from their feedback. Here we demonstrate how to use reinforcement learning from human feedback (RLHF) to improve upon simulated, embodied agents trained to a base level of competency with imitation learning. First, we collected data of humans interacting with agents in a simulated 3D world. We then asked annotators to record moments where they believed that agents either progressed toward or regressed from their human-instructed goal. Using this annotation data we leveraged a novel method - which we call "Inter-temporal Bradley-Terry" (IBT) modelling - to build a reward model that captures human judgments. Agents trained to optimise rewards delivered from IBT reward models improved with respect to all of our metrics, including subsequent human judgment during live interactions with agents. Altogether our results demonstrate how one can successfully leverage human judgments to improve agent behaviour, allowing us to use reinforcement learning in complex, embodied domains without programmatic reward functions. Videos of agent behaviour may be found at https://youtu.be/v_Z9F2_eKk4.

In this paper we present the implementation of a Control Barrier Function (CBF) using a quadratic program (QP) formulation that provides obstacle avoidance for a robotic manipulator arm system. CBF is a control technique that has emerged and developed over the past decade and has been extensively explored in the literature on its mathematical foundations, proof of set invariance and potential applications for a variety of safety-critical control systems. In this work we will look at the design of CBF for the robotic manipulator obstacle avoidance, discuss the selection of the CBF parameters and present a Reinforcement Learning (RL) scheme to assist with finding parameters values that provide the most efficient trajectory to successfully avoid different sized obstacles. We then create a data-set across a range of scenarios used to train a Neural-Network (NN) model that can be used within the control scheme to allow the system to efficiently adapt to different obstacle scenarios. Computer simulations (based on Matlab/Simulink) demonstrate the effectiveness of the proposed algorithm.

Deep reinforcement learning (DRL) is one of the most popular algorithms to realize an autonomous driving (AD) system. The key success factor of DRL is that it embraces the perception capability of deep neural networks which, however, have been proven vulnerable to Trojan attacks. Trojan attacks have been widely explored in supervised learning (SL) tasks (e.g., image classification), but rarely in sequential decision-making tasks solved by DRL. Hence, in this paper, we explore Trojan attacks on DRL for AD tasks. First, we propose a spatio-temporal DRL algorithm based on the recurrent neural network and attention mechanism to prove that capturing spatio-temporal traffic features is the key factor to the effectiveness and safety of a DRL-augment AD system. We then design a spatial-temporal Trojan attack on DRL policies, where the trigger is hidden in a sequence of spatial and temporal traffic features, rather than a single instant state used in existing Trojan on SL and DRL tasks. With our Trojan, the adversary acts as a surrounding normal vehicle and can trigger attacks via specific spatial-temporal driving behaviors, rather than physical or wireless access. Through extensive experiments, we show that while capturing spatio-temporal traffic features can improve the performance of DRL for different AD tasks, they suffer from Trojan attacks since our designed Trojan shows high stealthy (various spatio-temporal trigger patterns), effective (less than 3.1\% performance variance rate and more than 98.5\% attack success rate), and sustainable to existing advanced defenses.