Autonomous Robots最新文献

Humanoid robots have strong adaptability to complex environments and possess human-like flexibility, enabling them to perform precise farming and harvesting tasks in varying depths of terrains. They serve as essential tools for agricultural intelligence. In this article, a novel method was proposed to improve the robustness of autonomous navigation for humanoid robots, which intercommunicates the data fusion of the footprint planning and control levels. In particular, a deep reinforcement learning model - Proximal Policy Optimization (PPO) that has been fine-tuned is introduced into this layer, before which heuristic trajectory was generated based on imitation learning. In the RL period, the KL divergence between the agent’s policy and imitative expert policy as a value penalty is added to the advantage function. As a proof of concept, our navigation policy is trained in a robotic simulator and then successfully applied to the physical robot GTX for indoor multi-mode navigation. The experimental results conclude that incorporating imitation learning imparts anthropomorphic attributes to robots and facilitates the generation of seamless footstep patterns. There is a significant improvement in ZMP trajectory in y-direction from the center by 21.56% is noticed. Additionally, this method improves dynamic locomotion stability, the body attitude angle falling between less than ± 5.5(^circ ) compared to ± 48.4(^circ ) with traditional algorithm. In general, navigation error is below 5 cm, which we verified in the experiments. It is thought that the outcome of the proposed framework presented in this article can provide a reference for researchers studying autonomous navigation applications of humanoid robots on uneven ground.

Terrain traversability estimation is a fundamental task for supporting robot navigation on uneven surfaces. Recent learning-based approaches for predicting traversability from RGB images have shown promising results, but require manual annotation of a large number of images for training. To address this limitation, we present a method for traversability estimation on unlabeled videos that combines dataset synthesis, self-supervision and unsupervised domain adaptation. We pose the traversability estimation as a vector regression task over vertical bands of the observed frame. The model is pre-trained through self-supervision to reduce the distribution shift between synthetic and real data and encourage shared feature learning. Then, supervised training on synthetic videos is carried out, while employing an unsupervised domain adaptation loss to improve its generalization capabilities on real scenes. Experimental results show that our approach is on par with standard supervised training, and effectively supports robot navigation without the need of manual annotations. Training code and synthetic dataset will be publicly released at: https://github.com/perceivelab/traversability-synth.

Spatial search, and environmental monitoring are key technologies in robotics. These problems can be reformulated as maximal coverage problems with routing constraints, which are NP-hard problems. The generalized cost-benefit algorithm (GCB) can solve these problems with theoretical guarantees. To achieve better performance, evolutionary algorithms (EA) boost its performance via more samples. However, it is hard to know the terminal conditions of EA to outperform GCB. To solve these problems with theoretical guarantees and terminal conditions, in this research, the cross-entropy based Monte Carlo Tree Search algorithm (CE-MCTS) is proposed. It consists of three parts: the EA for sampling the branches, the upper confidence bound policy for selections, and the estimation of distribution algorithm for simulations. The experiments demonstrate that the CE-MCTS outperforms benchmark approaches (e.g., GCB, EAMC) in spatial search problems.

It is often unnoticed that the predominant way to use collocation methods is fundamentally flawed when applied to optimal control in robotics. Such methods assume that the system dynamics is given by a first order ODE, whereas robots are often governed by a second or higher order ODE involving configuration variables and their time derivatives. To apply a collocation method, therefore, the usual practice is to resort to the well known procedure of casting an Mth order ODE into M first order ones. This manipulation, which in the continuous domain is perfectly valid, leads to inconsistencies when the problem is discretized. Since the configuration variables and their time derivatives are approximated with polynomials of the same degree, their differential dependencies cannot be fulfilled, and the actual dynamics is not satisfied, not even at the collocation points. This paper draws attention to this problem, and develops improved versions of the trapezoidal and Hermite–Simpson collocation methods that do not present these inconsistencies. In many cases, the new methods reduce the dynamics transcription error in one order of magnitude, or even more, without noticeably increasing the cost of computing the solutions.

Mobile service robots are a promising technology for supporting workflows throughout the hospital. Combined with an understanding of the environment and the current situation, such systems have the potential to become invaluable tools for overcoming personal shortages and streamlining healthcare workflows. However, few robotic systems have actually been translated to practical application so far, which is due to many challenges centered around the strict and unique requirements imposed by the different hospital environments, which have not yet been collected and analyzed in a structured manner. To address this need, we now present a comprehensive classification of different dimensions of risk to be considered when designing mobile service robots for the hospital. Our classification consists of six risk categories – environmental complexity, hygienic requirements, interaction with persons and objects, workflow flexibility and autonomy – for each of which a scale with distinct risk levels is provided. This concept, for the first time allows for a precise classification of mobile service robots for the hospital, which can prove useful for certification and admission procedures as well as for defining architectural and safety requirements throughout the design process of such robots.

This paper presents the methods used by team CSIRO Data61 for multi-agent coordination and exploration in the DARPA Subterranean (SubT) Challenge. The SubT competition involved a single operator sending teams of robots to rapidly explore underground environments with severe navigation and communication challenges. Coordination was framed as a multi-robot task allocation (MRTA) problem to allow for a seamless integration of exploration with other required tasks. Methods for extending a consensus-based task allocation approach for an online and highly dynamic mission are discussed. Exploration tasks were generated from frontiers in a map of traversable space, and graph-based heuristics applied to guide the selection of exploration tasks. Results from simulation, field testing, and the final competition are presented. Team CSIRO Data61 tied for most points scored and achieved second place during the final SubT event.

For a robot to personalize physical assistance effectively, it must learn user preferences that can be generally reapplied to future scenarios. In this work, we investigate personalization of household cleanup with robots that can tidy up rooms by picking up objects and putting them away. A key challenge is determining the proper place to put each object, as people’s preferences can vary greatly depending on personal taste or cultural background. For instance, one person may prefer storing shirts in the drawer, while another may prefer them on the shelf. We aim to build systems that can learn such preferences from just a handful of examples via prior interactions with a particular person. We show that robots can combine language-based planning and perception with the few-shot summarization capabilities of large language models to infer generalized user preferences that are broadly applicable to future interactions. This approach enables fast adaptation and achieves 91.2% accuracy on unseen objects in our benchmark dataset. We also demonstrate our approach on a real-world mobile manipulator called TidyBot, which successfully puts away 85.0% of objects in real-world test scenarios.

When robots perform long action sequences, users will want to easily and reliably find out what they have done. We therefore demonstrate the task of learning to summarize and answer questions about a robot agent’s past actions using natural language alone. A single system with a large language model at its core is trained to both summarize and answer questions about action sequences given ego-centric video frames of a virtual robot and a question prompt. To enable training of question answering, we develop a method to automatically generate English-language questions and answers about objects, actions, and the temporal order in which actions occurred during episodes of robot action in the virtual environment. Training one model to both summarize and answer questions enables zero-shot transfer of representations of objects learned through question answering to improved action summarization.

We propose Text2Motion, a language-based planning framework enabling robots to solve sequential manipulation tasks that require long-horizon reasoning. Given a natural language instruction, our framework constructs both a task- and motion-level plan that is verified to reach inferred symbolic goals. Text2Motion uses feasibility heuristics encoded in Q-functions of a library of skills to guide task planning with Large Language Models. Whereas previous language-based planners only consider the feasibility of individual skills, Text2Motion actively resolves geometric dependencies spanning skill sequences by performing geometric feasibility planning during its search. We evaluate our method on a suite of problems that require long-horizon reasoning, interpretation of abstract goals, and handling of partial affordance perception. Our experiments show that Text2Motion can solve these challenging problems with a success rate of 82%, while prior state-of-the-art language-based planning methods only achieve 13%. Text2Motion thus provides promising generalization characteristics to semantically diverse sequential manipulation tasks with geometric dependencies between skills. Qualitative results are made available at https://sites.google.com/stanford.edu/text2motion.