IROS 2023 论文列表

Many state-of-the-art robotic applications utilize series elastic actuators (SEAs) with closed-loop force control to achieve complex tasks such as walking, lifting, and manipulation. Model-free PID control methods are more prone to instability due to nonlinearities in the SEA where cascaded model-based robust controllers can remove these effects to achieve stable force control. However, these model-based methods require detailed investigations to characterize the system accurately. Deep reinforcement learning (DRL) has proved to be an effective model-free method for continuous control tasks, where few works deal with hardware learning. This paper describes the training process of a DRL policy on the hardware of an SEA pendulum system for tracking force control trajectories from 0.05 - 0.35 Hz at 50 N amplitude using the Proximal Policy Optimization (PPO) algorithm. Safety mechanisms are developed and utilized for training the policy for over 21 hours (including overnight) without an operator present. The tracking performance is evaluated showing improvements of 25 N in mean absolute error when comparing the first 18 minutes of training to the full 21 hours for a 50 N amplitude, 0.1 Hz sinusoid desired force trajectory. Finally, the DRL policy exhibits better tracking and stability margins when compared to a model-free PID controller for a 50 N chirp force trajectory.

The human world is full of risks that threaten failure of robotic tasks. Dynamic robots, such as agile drones and walking bipeds, are particularly susceptible to failure because their time to make critical decisions is short. This work seeks a control algorithm which adapts to failures and reprioritizes robot behavior automatically, all at real-time speeds. Our failure-adaptive control framework learns failure probabilities from in situ experience and minimizes the risk of future failures using fast online planners (i.e. model predictive control). By reasoning about probabilities of failure, more imminent risks are automatically prioritized by the framework without manually tuning weighting factors. Further, our low-order probability model is learned using fast convex optimizations, allowing for immediate learning from triggered failures during operation. We demonstrate the framework's capability to learn and plan in real time (< 20 ms) in highly dynamic scenarios with micro-aerial vehicles (i.e. drones). We conduct two experiments: a chase-avoid task, and a chase-avoid - track task. In both scenarios, a single failure causes a categorical shift in robot behavior and the drone will adapt, plan, and execute a non-failing strategy within one second post-failure.

Unmanned ground vehicles (UGVs) require effective perception and analysis of their surrounding terrain for safe operation. This paper presents a novel approach to their local elevation mapping and traversability analysis using sparse data from a single LiDAR sensor, which can generate a dense local traversability map in real-time. By preserving ground height information, our method can differentiate between vertical obstacles, suspended objects and other terrains in the elevation map. The modified Bayesian generalized kernel elevation inference is utilized to predict and fill in sparse elevation maps to achieve local dense terrain traversability mapping. The system uses GPU parallel processing to accelerate calculations, ensuring real-time perception and dynamic processing. The proposed system was tested in both structured and unstructured environments, and achieved better performances in map filling and handling of suspended and vertical objects compared to other existing approaches.

Simulation systems of robots can facilitate the prediction, development, and debugging of robotic systems. However, they seldom applied in robotics education for primary and secondary school students. In this paper, we present a sim-to-real robotic system that enables students to optimize their algorithms in a simulated environment and validate them in a remote physical laboratory with data logs and remote cameras. Moreover, the system employs an automated submit-test-reset subsystem that minimizes the need for human intervention and provides 24/7 testing support. Experimental data from a trial with 28 students in remote areas show that the sim-to-real robotic experimental environment has comparable learning outcomes to a pure real robot environment and is significantly better than a pure simulation environment. Given the results, we validate that our system can substantially reduce the costs of teaching equipment and space while maintaining high-quality robotics education.

How should a robot speak in a formal, quiet and dark, or a bright, lively and noisy environment? By designing robots to speak in a more social and ambient-appropriate manner we can improve perceived awareness and intelligence for these agents. We describe a process and results toward selecting robot voice styles for perceived social appropriateness and ambiance awareness. Understanding how humans adapt their voices in different acoustic settings can be challenging due to difficulties in voice capture in the wild. Our approach includes 3 steps: (a) Collecting and validating voice data interactions in virtual Zoom ambiances, (b) Exploration and clustering human vocal utterances to identify primary voice styles, and (c) Testing robot voice styles in recreated ambiances using projections, lighting and sound. We focus on food service scenarios as a proof-of-concept setting. We provide results using the Pepper robot's voice with different styles, towards robots that speak in a contextually appropriate and adaptive manner. Our results with N=120 participants provide evidence that the choice of voice style in different ambiances impacted a robot's perceived intelligence in several factors including: social appropriateness, comfort, awareness, human-likeness and competency.

We propose a co-navigation algorithm that enables a human and a robot to work together to navigate to a common goal. In this system, the human is responsible for making high-level steering decisions, and the robot, in turn, provides haptic feedback for collision avoidance and path suggestions while reacting to changes in the environment. Our algorithm uses optimized Rapidly-exploring Random Trees (RRT*) to generate paths to lead the user to the goal, via an attractive force feedback computed using a Control Lyapunov Function (CLF). We simultaneously ensure collision avoidance where necessary using a Control Barrier Function (CBF). We demonstrate our approach using simulations with a virtual pilot, and hardware experiments with a human pilot. Our results show that combining RRT* and CBFs is a promising tool for enabling collaborative human-robot navigation.

This paper proposes a reachability-aware model predictive control with a discrete control barrier function for backward obstacle avoidance for a tractor-trailer system. The framework incorporates the state-variant reachable set obtained through sampling-based reachability analysis and symbolic regression into the objective function of model predictive control. By optimizing the intersection of the reachable set and iterative non-safe region generated by the control barrier function, the system demonstrates better performance in terms of safety with a constant decay rate, while enhancing the feasibility of the optimization problem. The proposed algorithm improves real-time performance due to a shorter horizon and outperforms the state-of-the-art algorithms in the simulation environment and on a real robot.

Deep neural networks using for real-world classification task require high reliability and robustness. However, the Softmax output by the last layer of network is often over-confident. We propose a novel confidence estimation method by considering model quality for deep classification models. Two metrics, MQ-Repres and MQ-Discri are developed accordingly to evaluate the model quality, and also provide a new confidence estimation called MQ-Conf for online inference. We demonstrate the capability of the proposed method by the 3D3D semantic segmentation tasks using three different deep networks. Through confusion analysis and feature visualization we show the rationality and reliability of the model quality quantification method.11This work is supported by the National Natural Science Foundation of China under Grant U22A2061 and High-performance Computing Platform of Peking University.

Finger-tip tactile sensors are increasingly used for robotic sensing to establish stable grasps and to infer object properties. Promising performance has been shown in a number of works for inferring adjectives that describe the object, but there remains a question about how each taxel contributes to the performance. This paper explores this question with empirical experiments, leading insights for future finger-tip tactile sensor usage and design: one tactile sensor instead of a pair of sensors is sufficient for symmetric objects and interaction motions; dense taxels are beneficial for texture-related adjectives, but can be distracting to non-texture-related ones; and a frame-rate much lower than the BioTac sensor can satisfy the demand of inferring object adjectives in the PHAC-2 dataset.

Robot grasping is subject to an inherent tradeoff: Grippers with a large span typically take a longer time to close, and fast grippers usually cover a small span. However, many practical applications of grippers require the ability to close a large distance rapidly. For example, grasping cloth typically requires pressing a wide span of fabric into a graspable cusp. Besides, the ability to perform human-like grasping and ease offabrication are also very important for new soft grippers. Here, we demonstrate a human-finger-inspired snapping gripper that exploits elastic instability to achieve rapid and reversible closing over a wide span. Using prestressed semi-rigid material as the skeleton, the gripper fingers can widely open (86 mm) and rapidly close (46 ms) following a trajectory similar to that of a thumb-index finger pinching, and is 2.7 times and 10.9 times better than the reference gripper in terms of span and speed, respectively. We theoretically give the design principle, simulatively verify the method, and experimentally test this gripper on a variety of rigid, flexible, and limp objects and achieve good mechanical performance.

This work introduces range-based GP maps, which directly represent terrain by modeling the range from a LiDAR sensor as a Gaussian process (GP) in spherical space. Such a model aligns the predicted uncertainty from the GP regression with the uncertainty in the underlying sensor observations. Experimental evaluation on simulated natural terrain indicates that local range-based GP maps perform comparably to elevation-based methods when predicting terrain height, with the former producing more stable parameters and providing a better uncertainty representation. An aggregation method is proposed using the pose as an additional input to the GP. Unlike their elevation-based counterparts, range-based GP maps are capable of modeling overhangs and vertical obstacles with ease, demonstrated with examples of maps built on real-world data from a fully 3D subterranean environment.

This paper presents a new information source for supporting robot localisation: material composition. The proposed method complements the existing visual, structural, and semantic cues utilized in the literature. However, it has a distinct advantage in its ability to differentiate structurally [23], visually [25] or categorically [1] similar objects such as different doors, by using Raman spectrometers. Such devices can identify the material of objects it probes through the bonds between the material's molecules. Unlike similar sensors, such as mass spectroscopy, it does so without damaging the material or environment. In addition to introducing the first material-based localisation algorithm, this paper supports the future growth of the field by presenting a gazebo plugin for Raman spectrometers, material sensing demonstrations, as well as the first-ever localisation data-set with benchmarks for material-based localisation. This benchmarking shows that the proposed technique results in a significant improvement over current state-of-the-art localisation techniques, achieving 16 % more accurate localisation than the leading baseline. The code and dataset will be released at: https://github.com/ThirgoodC/RaSpectLoc

Due to the robustness in sensing, radar has been highlighted, overcoming harsh weather conditions such as fog and heavy snow. In this paper, we present a novel radar-only place recognition that measures the similarity score by utilizing Radon-transformed sinogram images and cross-correlation in frequency domain. Doing so achieves rigid transform invariance during place recognition, while ignoring the effects of radar multipath and ring noises. In addition, we compute the radar similarity distance using mutable threshold to mitigate variability of the similarity score, and reduce the time complexity of processing a copious radar data with hierarchical retrieval. We demonstrate the matching performance for both intra-session loop-closure detection and global place recognition using a publicly available imaging radar datasets. We verify reliable performance compared to existing stable radar place recognition method. Furthermore, codes for the proposed imaging radar place recognition is released for community https://github.com/hyesu-jang/RaPlace.

This paper presents RVWO, a system designed to provide robust localization and mapping for wheeled mobile robots in challenging scenarios. The proposed approach leverages a probabilistic framework that incorporates semantic prior information about landmarks and visual re-projection error to create a landmark reliability model, which acts as an adaptive kernel for the visual residuals in optimization. Additionally, we fuse visual residuals with wheel odometry measurements, taking advantage of the planar motion assumption. The RVWO system is designed to be robust against wrong data association due to moving objects, poor visual texture, bad illumination, and wheel slippage. Evaluation results demonstrate that the proposed system shows competitive results in dynamic environments and outperforms existing approaches on both public benchmarks and our custom hardware setup. We also provide the code as an open-source contribution to the robotics community22https://github.com/be2rlab/rvwo.

Household robots operate in the same space for years. Such robots incrementally build dynamic maps that can be used for tasks requiring remote object localization. However, benchmarks in robot learning often test generalization through inference on tasks in unobserved environments. In an observed environment, locating an object is reduced to choosing from among all object proposals in the environment, which may number in the 100,000s. Armed with this intuition, using only a generic vision-language scoring model with minor modifications for 3d encoding and operating in an embodied environment, we demonstrate an absolute performance gain of 9.84% on remote object grounding above state of the art models for REVERIE and of 5.04% on FAO. When allowed to pre-explore an environment, we also exceed the previous state of the art pre-exploration method on REVERIE. Additionally, we demonstrate our model on a real-world TurtleBot platform, highlighting the simplicity and usefulness of the approach. Our analysis outlines a “bag of tricks” essential for accomplishing this task, from utilizing 3d coordinates and context, to gener-alizing vision-language models to large 3d search spaces.

Reinforcement learning is used to tackle complex tasks with high-dimensional sensory inputs. Over the past decade, a wide range of reinforcement learning algorithms have been developed, with recent progress benefiting from deep learning for raw sensory signal representation. This raises a natural question: how well do these algorithms perform across different robotic manipulation tasks? To objectively compare algorithms, benchmarks use performance metrics. Benchmarks use objective performance metrics to offer a scientific way to compare algorithms. In this paper, we introduce RMBench, the first benchmark for robotic manipulations with high-dimensional continuous action and state spaces. We implement and evaluate reinforcement learning algorithms that take observed pixels as inputs and report their average performance and learning curves to demonstrate their performance and training stability. Our study concludes that none of the evaluated algorithms can handle all tasks well, with soft Actor-Critic outperforming most algorithms in terms of average reward and stability, and an algorithm combined with data augmentation potentially facilitating learning policies. Our code is publicly available at https://github.com/xiangyanfei212/RMBench-2022.git, including all benchmark tasks and studied algorithms.

A platoon refers to a group of vehicles traveling together in very close proximity using automated driving technology. Owing to its immense capacity to improve fuel efficiency, driving safety, and driver comfort, platooning technology has garnered substantial attention from the autonomous vehicle research community. Although highly advantageous, recent research has uncovered that an excessively small intra-platoon gap can impede traffic flow during highway on-ramp merging. While existing control-based methods allow for adaptation of the intra-platoon gap to improve traffic flow, making an optimal control decision under the complex dynamics of traffic conditions remains a challenge due to the massive computational complexity. In this paper, we present the design, implementation, and evaluation of a novel reinforcement learning framework that adaptively adjusts the intra-platoon gap of an individual platoon member to maximize traffic flow in response to dynamically changing, complex traffic conditions for highway on-ramp merging. The framework's state space has been meticulously designed in consultation with the transportation literature to take into account critical traffic parameters that bear direct relevance to merging efficiency. An intra-platoon gap decision making method based on the deep deterministic policy gradient algorithm is created to incorporate the continuous action space to ensure precise and continuous adaptation of the intra-platoon gap. An extensive simulation study demonstrates the effectiveness of the reinforcement learning-based approach for significantly improving traffic flow in various highway on-ramp merging scenarios.

This paper proposes a novel approach to address the technical challenges of stable object grasping, particularly in the context of handling tableware in a home environment. Handling tableware is particularly important, yet challenging, due to the flat nature of most tableware objects and the need to maintain a stable posture to prevent spills. To address these challenges, we present three key contributions: 1) a large-scale tableware dataset, not commonly found in the previous datasets; 2) a novel sampling method for stable grasp pose generation; and 3) a multi-modal fusion grasp network that effectively learns 6- DoF grasp pose, including flat objects. Our dataset contains over 45 million grasp poses and 1 million RGBD images captured in 800 scenes, which include randomly selected 10–18 tableware objects under 4 different lighting conditions. The grasp poses in the dataset are generated using a novel sampling method that incorporates geometric analysis to ensure stable grasping with minimal object movement. Furthermore, we design an RGBD fusion grasp network (RGBD-FGN) that can combine information from RGB and depth images considering each characteristic. Our experimental results demonstrate the superior performance of our approach over existing techniques, which is a significant contribution towards developing a multitasking home robot. Our dataset and source code can be accessed at https://github.com/SamsungLabs/RGBD-FGN.

High-performance real-time 3D object detection is crucial in autonomous driving perception systems. Voxel-or point-based 3D object detectors are highly accurate but inefficient and difficult to deploy, while other methods use 2D projection views to improve efficiency, but information loss usually degrades performance. To balance effectiveness and efficiency, we propose a scheme called RFDNet that uses range features to decorate points. Specifically, RFDNet adaptively aggregates point features projected to independent grids and nearby regions via Dilated Grid Feature Encoding (DGFE) to generate a range view, which can handle occlusion and multi-frame inputs while the established geometric correlation between grid with surrounding space weakens the effects of scale distortion. We also propose a Soft Box Regression (SBR) strategy that supervises 3D box regression on a more extensive range than conventional methods to enhance model robustness. In addition, RFDNet benefits from our designed Semantic-assisted Ground-truth Sample (SA-GTS) data augmentation, which additionally considers collisions and spatial distributions of objects. Experiments on the nuScenes benchmark show that RFDNet outperforms all LiDAR-only non-ensemble 3D object detectors and runs at high speed of 20 FPS, achieving a better effectiveness-efficiency trade-off. Code is available at https://github.com/wy17646051/RFDNet.