Abstract:Large language models (LLMs) have shown significant potential in guiding embodied agents to execute language instructions across a range of tasks, including robotic manipulation and navigation. However, existing methods are primarily designed for static environments and do not leverage the agent's own experiences to refine its initial plans. Given that real-world environments are inherently stochastic, initial plans based solely on LLMs' general knowledge may fail to achieve their objectives, unlike in static scenarios. To address this limitation, this study introduces the Experience-and-Emotion Map (E2Map), which integrates not only LLM knowledge but also the agent's real-world experiences, drawing inspiration from human emotional responses. The proposed methodology enables one-shot behavior adjustments by updating the E2Map based on the agent's experiences. Our evaluation in stochastic navigation environments, including both simulations and real-world scenarios, demonstrates that the proposed method significantly enhances performance in stochastic environments compared to existing LLM-based approaches. Code and supplementary materials are available at https://e2map.github.io/.
Abstract:Reinforcement learning (RL) provides a compelling framework for enabling autonomous vehicles to continue to learn and improve diverse driving behaviors on their own. However, training real-world autonomous vehicles with current RL algorithms presents several challenges. One critical challenge, often overlooked in these algorithms, is the need to reset a driving environment between every episode. While resetting an environment after each episode is trivial in simulated settings, it demands significant human intervention in the real world. In this paper, we introduce a novel autonomous algorithm that allows off-the-shelf RL algorithms to train an autonomous vehicle with minimal human intervention. Our algorithm takes into account the learning progress of the autonomous vehicle to determine when to abort episodes before it enters unsafe states and where to reset it for subsequent episodes in order to gather informative transitions. The learning progress is estimated based on the novelty of both current and future states. We also take advantage of rule-based autonomous driving algorithms to safely reset an autonomous vehicle to an initial state. We evaluate our algorithm against baselines on diverse urban driving tasks. The experimental results show that our algorithm is task-agnostic and achieves better driving performance with fewer manual resets than baselines.
Abstract:Autonomous navigation in extreme mountainous terrains poses challenges due to the presence of mobility-stressing elements and undulating surfaces, making it particularly difficult compared to conventional off-road driving scenarios. In such environments, estimating traversability solely based on exteroceptive sensors often leads to the inability to reach the goal due to a high prevalence of non-traversable areas. In this paper, we consider traversability as a relative value that integrates the robot's internal state, such as speed and torque to exhibit resilient behavior to reach its goal successfully. We separate traversability into apparent traversability and relative traversability, then incorporate these distinctions in the optimization process of sampling-based planning and motion predictive control. Our method enables the robots to execute the desired behaviors more accurately while avoiding hazardous regions and getting stuck. Experiments conducted on simulation with 27 diverse types of mountainous terrain and real-world demonstrate the robustness of the proposed framework, with increasingly better performance observed in more complex environments.
Abstract:Autonomous vehicles have been actively investigated over the past few decades. Several recent works show the potential of autonomous driving transportation services in urban environments with impressive experimental results. However, these works note that autonomous vehicles are still occasionally inferior to expert drivers in complex scenarios. Furthermore, they do not focus on the possibilities of autonomous driving transportation services in other areas beyond urban environments. This paper presents the research results and lessons learned from autonomous driving transportation services in airfield, crowded indoor, and urban environments. We discuss how we address several unique challenges in these diverse environments. We also offer an overview of remaining challenges that have not received much attention but must be addressed. This paper aims to share our unique experience to support researchers who are interested in realizing the potential of autonomous vehicles in various real-world environments.
Abstract:In this study, we address the off-road traversability estimation problem, that predicts areas where a robot can navigate in off-road environments. An off-road environment is an unstructured environment comprising a combination of traversable and non-traversable spaces, which presents a challenge for estimating traversability. This study highlights three primary factors that affect a robot's traversability in an off-road environment: surface slope, semantic information, and robot platform. We present two strategies for estimating traversability, using a guide filter network (GFN) and footprint supervision module (FSM). The first strategy involves building a novel GFN using a newly designed guide filter layer. The GFN interprets the surface and semantic information from the input data and integrates them to extract features optimized for traversability estimation. The second strategy involves developing an FSM, which is a self-supervision module that utilizes the path traversed by the robot in pre-driving, also known as a footprint. This enables the prediction of traversability that reflects the characteristics of the robot platform. Based on these two strategies, the proposed method overcomes the limitations of existing methods, which require laborious human supervision and lack scalability. Extensive experiments in diverse conditions, including automobiles and unmanned ground vehicles, herbfields, woodlands, and farmlands, demonstrate that the proposed method is compatible for various robot platforms and adaptable to a range of terrains. Code is available at https://github.com/yurimjeon1892/FtFoot.
Abstract:We address the challenge of estimating the learning rate for adaptive gradient methods used in training deep neural networks. While several learning-rate-free approaches have been proposed, they are typically tailored for steepest descent. However, although steepest descent methods offer an intuitive approach to finding minima, many deep learning applications require adaptive gradient methods to achieve faster convergence. In this paper, we interpret adaptive gradient methods as steepest descent applied on parameter-scaled networks, proposing learning-rate-free adaptive gradient methods. Experimental results verify the effectiveness of this approach, demonstrating comparable performance to hand-tuned learning rates across various scenarios. This work extends the applicability of learning-rate-free methods, enhancing training with adaptive gradient methods.
Abstract:Hierarchical reinforcement learning (HRL) has led to remarkable achievements in diverse fields. However, existing HRL algorithms still cannot be applied to real-world navigation tasks. These tasks require an agent to perform safety-aware behaviors and interact with surrounding objects in dynamic environments. In addition, an agent in these tasks should perform consistent and structured exploration as they are long-horizon and have complex structures with diverse objects and task-specific rules. Designing HRL agents that can handle these challenges in real-world navigation tasks is an open problem. In this paper, we propose imagination-augmented HRL (IAHRL), a new and general navigation algorithm that allows an agent to learn safe and interactive behaviors in real-world navigation tasks. Our key idea is to train a hierarchical agent in which a high-level policy infers interactions by interpreting behaviors imagined with low-level policies. Specifically, the high-level policy is designed with a permutation-invariant attention mechanism to determine which low-level policy generates the most interactive behavior, and the low-level policies are implemented with an optimization-based behavior planner to generate safe and structured behaviors following task-specific rules. To evaluate our algorithm, we introduce five complex urban driving tasks, which are among the most challenging real-world navigation tasks. The experimental results indicate that our hierarchical agent performs safety-aware behaviors and properly interacts with surrounding vehicles, achieving higher success rates and lower average episode steps than baselines in urban driving tasks.
Abstract:A significant bottleneck in applying current reinforcement learning algorithms to real-world scenarios is the need to reset the environment between every episode. This reset process demands substantial human intervention, making it difficult for the agent to learn continuously and autonomously. Several recent works have introduced autonomous reinforcement learning (ARL) algorithms that generate curricula for jointly training reset and forward policies. While their curricula can reduce the number of required manual resets by taking into account the agent's learning progress, they rely on task-specific knowledge, such as predefined initial states or reset reward functions. In this paper, we propose a novel ARL algorithm that can generate a curriculum adaptive to the agent's learning progress without task-specific knowledge. Our curriculum empowers the agent to autonomously reset to diverse and informative initial states. To achieve this, we introduce a success discriminator that estimates the success probability from each initial state when the agent follows the forward policy. The success discriminator is trained with relabeled transitions in a self-supervised manner. Our experimental results demonstrate that our ARL algorithm can generate an adaptive curriculum and enable the agent to efficiently bootstrap to solve sparse-reward maze navigation tasks, outperforming baselines with significantly fewer manual resets.
Abstract:Robotic agents trained using reinforcement learning have the problem of taking unreliable actions in an out-of-distribution (OOD) state. Agents can easily become OOD in real-world environments because it is almost impossible for them to visit and learn the entire state space during training. Unfortunately, unreliable actions do not ensure that agents perform their original tasks successfully. Therefore, agents should be able to recognize whether they are in OOD states and learn how to return to the learned state distribution rather than continue to take unreliable actions. In this study, we propose a novel method for retraining agents to recover from OOD situations in a self-supervised manner when they fall into OOD states. Our in-depth experimental results demonstrate that our method substantially improves the agent's ability to recover from OOD situations in terms of sample efficiency and restoration of the performance for the original tasks. Moreover, we show that our method can retrain the agent to recover from OOD situations even when in-distribution states are difficult to visit through exploration.
Abstract:Although contrastive learning methods have shown prevailing performance on a variety of representation learning tasks, they encounter difficulty when the training dataset is long-tailed. Many researchers have combined contrastive learning and a logit adjustment technique to address this problem, but the combinations are done ad-hoc and a theoretical background has not yet been provided. The goal of this paper is to provide the background and further improve the performance. First, we show that the fundamental reason contrastive learning methods struggle with long-tailed tasks is that they try to maximize the mutual information maximization between latent features and input data. As ground-truth labels are not considered in the maximization, they are not able to address imbalances between class labels. Rather, we interpret the long-tailed recognition task as a mutual information maximization between latent features and ground-truth labels. This approach integrates contrastive learning and logit adjustment seamlessly to derive a loss function that shows state-of-the-art performance on long-tailed recognition benchmarks. It also demonstrates its efficacy in image segmentation tasks, verifying its versatility beyond image classification.