IROS 2022 Papers

This article presents an entirely data-driven approach for autonomous control of redundant manipulators with hydraulic actuation. The approach only requires minimal system information, which is inherited from a simulation model. The non-linear hydraulic actuation dynamics are modeled using actuator networks from the data gathered during the manual operation of the manipulator to effectively emulate the real system in a simulation environment. A neural network control policy for autonomous control, based on end-effector (EE) position tracking is then learned using Reinforcement Learning (RL) with Ornstein-Uhlenbeck process noise (OUNoise) for efficient exploration. The RL agent also receives feedback based on supervised learning of the forward kinematics which facilitates selecting the best suitable action from exploration. The control policy directly provides the joint variables as outputs based on provided target EE position while taking into account the system dynamics. The joint variables are then mapped to the hydraulic valve commands, which are then fed to the system without further modifications. The proposed approach is implemented on a scaled hydraulic forwarder crane with three revolute and one prismatic joint to track the desired position of the EE in 3-Dimensional (3D) space. With the emulated dynamics and extensive learning in simulation, the results demonstrate the feasibility of deploying the learned controller directly on the real system.

This paper deals with the control of a laser spot in the context of minimally invasive surgery of the middle ear, e.g., cholesteatoma removal. More precisely, our work is concerned with the exhaustive burring of residual infected cells after primary mechanical resection of the pathological tissues since the latter cannot guarantee the treatment of all the infected tissues, the remaining infected cells cause regeneration of the diseases in 20%-25 % of cases, which require a second surgery 12–18 months later. To tackle such a complex surgery, we have developed a robotic platform that consists of the combination of a macro-scale system (7 degrees of freedoms (DoFs) robotic arm) and a micro-scale flexible system (2 DoFs) which operates inside the middle ear cavity. To be able to treat the residual cholesteatoma regions, we proposed a method to generate optimal laser automatically scanning trajectories inside the areas and between them. The trajectories are tacked using an image-based control scheme. The proposed method and materials were validated experimentally using the lab-made robotic platform. The obtained results in terms of accuracy and behaviour meet the laser surgery requirements perfectly.

Designing controllers for complex robots such as humanoids is not an easy task. Often, researchers hand-tune controllers, but this is a time-consuming approach that yields a single controller which cannot generalize well to varied tasks. This work presents a method which uses the NSGA-II multi-objective optimization algorithm with various training trajectories to output a diverse Pareto set of well-functioning controller weights and gains. The best of these are shown to also work well on the real Talos robot. The learned Pareto front is then used in a Bayesian optimization (BO) algorithm both as a search space and as a source of prior information in the initial mean estimate. This combined learning approach, leveraging the two optimization methods together, finds a suitable parameter set for a new trajectory within 20 trials and outperforms both BO in the continuous parameter search space and random search along the precomputed Pareto front. The few trials required for this formulation of BO suggest that it could feasibly be applied on the physical robot using a Pareto front generated in simulation.

Online trajectory planners enable quadrotors to safely and smoothly navigate in unknown cluttered environments. However, tuning parameters is challenging since modern planners have become too complex to mathematically model and predict their interaction with unstructured environments. This work takes humans out of the loop by proposing a planner parameter adaptation framework that formulates objectives into two complementary categories and optimizes them asynchronously. Objectives evaluated with and without trajectory execution are optimized using Bayesian Optimization (BayesOpt) and Particle Swarm Optimization (PSO), respectively. By combining two kinds of objectives, the total convergence rate of the black-box optimization is accelerated while the dimension of optimized parameters can be increased. Benchmark comparisons demonstrate its superior performance over other strategies. Tests with changing obstacle densities validate its real-time environment adaption, which is difficult for prior manual tuning. Real-world flights with different drone platforms, environments, and planners show the proposed framework's scalability and effectiveness.

Robot-Assisted Minimally Invasive Surgery (RAMIS), which introduced robot-actuated invasive tools to increase the dexterity and efficiency of traditional MIS, has become popular. Investigations on how to achieve autonomy in RAMIS have drawn vast intention recently, which urges further insights into the process of the surgical procedures. In this paper, the definition of critical actions, which discriminates the essential stages from regular surgical actions, is proposed to help decompose the complicated surgical processes. A critical intra-operative moment of the surgical workflow, which is called the keyframe, is introduced to indicate the beginning or ending moments of the critical actions. A keyframe detection method is proposed for critical action identification based on a new in-vivo dataset labeled by expert surgeons. Surgeons' criteria for critical actions are captured by the explainable features, which can be extracted from the raw laparoscopic images with a two-stage network. Motivated by the surgeon's decision process of keyframes, a hierarchical structure is designed for keyframe identification by checking the spatial-temporal characteristics of the explainable features. Experimental results show that the reliability of the proposed method for keyframe detection achieves unanimous agreement by expert surgeons.

We propose an optimization system for motion planning of robot arms using Petri Nets. The proposed optimization system consists of four sub-systems consisting of automatic generation of Petri Nets from event log data, optimization system of firing sequence of derived Petri Net model, verification system using Petri Net simulation, and an automatic program generation system. The model generation system automatically generates the Petri Net model from the event logs using process mining. The Petri Net verification system is used to check the consistency of the generated Petri Nets to obtain the optimal firing sequence for robot motion. The motion planning algorithm generates motion programs for robots based on optimal firing sequences. The proposed optimization model is applied to a 6-DOF (Degree of Freedom) robot manipulator (Niryo Ned). Experimental results show that the proposed method achieves motion plan optimization for the pick-and-place operation with different robot configurations.

Unmanned aerial vehicles (UAVs) have broad applications including disaster response, transportation, photography, and mapping. A significant bottleneck in the development of UAVs is the limited availability of automatic tools for task-specific co-design of a UAV's shape and controller. The development of such tools is particularly challenging as UAVs can take many forms, including fixed-wing planes, radial copters, and hybrid topologies, with each class of topology showing different advantages. In this work, we present a computational design pipeline for UAVs based on a graph grammar that can search across a wide range of topologies. Graphs generated by the grammar encode different topologies and component selections, while continuous parameters encode the dimensions and properties of each component. We further augment the shape representation with deformation cages, which allow expressing a variety of wing shapes. Each UAV design is associated with an LQR controller with tunable continuous parameters. To search over this complex discrete and continuous design space, we develop a hybrid algorithm that combines discrete graph search strategies and gradient-based continuous optimization methods using a differentiable UAV simulator. We evaluate our pipeline on a set of simulated flight tasks requiring dynamic motions, showing that it discovers novel UAV designs that outperform canonical UAVs typically made by engineers.

Soft robotic grippers are becoming increasingly popular for agricultural and logistics automation. Their passive conformability enables them to adapt to varying product shapes and sizes, providing stable large-area grasps. This work presents a novel methodology for combining soft robotic grippers with electrical impedance tomography-based sensors to infer intrinsic properties of grasped fruits. We use a Fin Ray soft robotic finger with embedded microspines to grab and obtain rich multi-direction electrical properties of the object. Learning-based techniques are then used to infer the desired fruit properties. The framework is extensively tested and validated on multiple fruit groups. Our results show that ripeness parameters and even weight of the grasped fruit can be estimated with reasonable accuracy autonomously using the proposed system.

Planning a safe trajectory for minimally invasive (keyhole) neurosurgery procedures require avoiding critical anatomical structures such as blood vessels and ventricles while optimizing the needle trajectory parameters such as length and curvature to comply with the needle kinematics. In this paper, we propose a reinforcement learning-based method for obtaining kinematically feasible trajectories for flexible needle insertions. Proposed approach utilizes Bezier curve control points that are generated by a reward-based reinforcement learning framework, called Flexible Needle Path Generation (FNPG). FNPG framework is trained using an environment that consists of (1) critical structures (e.g. ventricles) obtained through atlas based segmentation of MRI-T1 images, (2) blood vessels segmented from MR angiography (MRA) data and, (3) simulated brain tumor with varying size and location. The curvilinear paths obtained through the FNPG framework are compared with the traditional sampling based algorithm RRT*. The results show that the FNPG approach can produce smoother and shorter trajectories compared to RRT* while avoiding the critical anatomical structures.

The tasks that unmanned aerial vehicles (UAVs) have taken upon have progressively grown in complexity over the years, alongside with the level of autonomy with which they are carried out. In this work, we present an example of aerial screwing operations with a fully-actuated tilt-rotor platform. Key contributions include a new control framework to automate screwing operations through a robust hole search and in-hole detection algorithm. These are achieved without a-priori knowledge of the exact hole location, and without the use of external tools, such as vision based hole detection or force sensors. Wrench coupling is implemented to account for the platform's kinematic constraints during screwing. The application of a constant contact force and a compliant response to induced disturbances are obtained with the use of admittance control. The full framework is validated with extensive flight experiments that demonstrate the effectiveness of each subsystem, as well as the complete architecture. We also validate the robustness of the detection algorithm against false positives. Within the results we demonstrate the ability to perform the automated task with a 86% success rate over 35 flights, and measured hole search time of 9s (median value).

We propose a fully autonomous system for mobile robot exploration in unknown environments. Our system employs a novel frontier detection algorithm based on the fast front propagation (FFP) technique and uses parallel path planning to reach the detected front regions. Given an occupancy grid map in 2D, possibly updated online, our algorithm can find all the frontier points that can allow mobile robots to visit unexplored regions to maximize the exploratory coverage. Our FFP method is six~seven times faster than the state-of-the-art wavefront frontier detection algorithm in terms of finding frontier points without compromising the detection accuracy. The speedup can be further accelerated by simplifying the map without degrading the detection accuracy. To expedite locating the optimal frontier point, We also eliminate spurious points by the obstacle filter and the novel boundary filter. In addition, we parallelize the global planning phase using the branch-and-bound A*, where the search space of each thread is confined by its best knowledge discovered during the parallel search. As a result, our parallel path-planning algorithm operating on 20 threads is about 30 times faster than the vanilla exploration system that operates on a single thread. Our method is validated through extensive experiments, including autonomous robot exploration in both synthetic and real-world scenarios. In the real-world experiment, we show that an autonomous navigation system using a human-sized mobile manipulator robot equipped with a low-end embedded processor that fully integrates our FFP and parallel path-planning algorithms.

Visual action planning particularly excels in applications where the state of the system cannot be computed explicitly, such as manipulation of deformable objects, as it enables planning directly from raw images. Even though the field has been significantly accelerated by deep learning techniques, a crucial requirement for their success is the availability of a large amount of data. In this work, we propose the Augment-Connect-Explore (ACE) paradigm to enable visual action planning in cases of data scarcity. We build upon the Latent Space Roadmap (LSR) framework which performs planning with a graph built in a low dimensional latent space. In particular, ACE is used to i) Augment the available training dataset by autonomously creating new pairs of datapoints, ii) create new unobserved Connections among representations of states in the latent graph, and iii) Explore new regions of the latent space in a targeted manner. We validate the proposed approach on both simulated box stacking and real-world folding task showing the applicability for rigid and deformable object manipulation tasks, respectively.

Reflective and textureless surfaces such as windows, mirrors, and walls can be a challenge for scene reconstruction, due to depth discontinuities and holes. We propose an audio-visual method that uses the reflections of sound to aid in depth estimation and material classification for 3D scene reconstruction in robot navigation and AR/VR applications. The mobile phone prototype emits pulsed audio, while recording video for audio-visual classification for 3D scene reconstruction. Reflected sound and images from the video are input into our audio (EchoCNN-A) and audio-visual (EchoCNN-AV) convolutional neural networks for surface and sound source detection, depth estimation, and material classification. The inferences from these classifications enhance 3D scene reconstructions containing open spaces and reflective surfaces by depth filtering, inpainting, and placement of unmixed sound sources in the scene. Our prototype, demos, and experimental results from real-world with challenging surfaces and sound, also validated with virtual scenes, indicate high success rates on classification of material, depth estimation, and closed/open surfaces, leading to considerable improvement in 3D scene reconstruction for robot navigation.

This work focuses on estimating 6D poses and sizes of category-level objects from a single RGB-D image. How to exploit the complementary RGB and depth features plays an important role in this task yet remains an open question. Due to the large intra-category texture and shape variations, an object instance in test may have different RGB and depth features from those of the object instances in training, which poses challenges to previous RGB-D fusion methods. To deal with such problem, an Attention-guided RGB-D Fusion Network (ARF-Net) is proposed in this work. Our key design is an ARF module that learns to adaptively fuse RGB and depth features with guidance from both structure-aware attention and relation-aware attention. Specifically, the structure-aware attention captures spatial relationship among object parts and the relation-aware attention captures the RGB-to-depth correlations between the appearance and geometric features. Our ARF -Net directly establishes canonical correspondences with a compact decoder based on the multi-modal features from our ARF module. Extensive experiments show that our method can effectively fuse RGB features to various popular point cloud encoders and provide consistent performance improvement. In particular, without reconstructing instance 3D models, our method with its relatively compact architecture outperforms all state-of-the-art models on CAMERA25 and REAL275 benchmarks by a large margin.

A swarm of fixed-wing unmanned aerial vehicles (UAVs) is expected to efficiently accomplish various tasks in complex scenarios. This paper proposes an attention-based population-invariant multi-agent deep reinforcement learning (MADRL) approach to deal with the decentralized collision-free flocking problem for a scalable fixed-wing UAV swarm. First, this problem is modeled as a decentralized partially observable Markov decision process from the perspective of each follower. Then, an improved multi-agent deep deterministic policy gradient (MADDPG) algorithm is presented to efficiently learn the population-invariant flocking policy. In this algorithm, the parameter sharing with ego-centric representation mechanism is incorporated to improve learning efficiency. Besides, the attention-based population-invariant network structure (APINet) is designed by leveraging the self-attention mechanism. With this structure, the learned flocking policy is invariant to the population of the swarm. Finally, both numerical and hardware-in-the-loop simulation results verify the efficiency and scalability of the proposed approach.

Experienced human drivers always make safe driving decisions by selectively observing the front, rear and side- view mirrors. Several end - to-end methods have been pro-posed to learn driving models with multi-view visual infor-mation. However, these benchmark methods lack semantic understanding of multi-view image contents, where human drivers usually reason these information for decision making with different visual region of interests. In this paper, we propose an attention-based deep learning method to learn a driving model with input of surround-view visual information and the route planner, in which a multi-view attention module is designed for obtaining region of interests from human drivers. We evaluate our model on the Drive360 dataset with comparison of benchmarking deep driving models. Results demonstrate that our model achieves a competitive accuracy in both steering angle and speed prediction than benchmarking methods. Code is available at https://githuh.com/jet-uestc/MVA-Net.

We present a novel overconstraining and constraint-discarding method for asynchronous, real-time, decentralized, multi-robot trajectory planning that ensures collision avoidance. Our approach utilizes communication between robots. The communication medium is best-effort: messages may be dropped, re-ordered or delayed. Robots conservatively constrain themselves against others assuming they may be working with outdated information, and discard constraints when they receive update messages from others. Our method can augment existing synchronized decentralized receding horizon planning algorithms that utilize separating hyperplanes for collision avoidance thereby making them applicable to asynchronous setups. As an example, we extend an existing model predictive control based, synchronized, decentralized multi-robot planner using our method. We show our method's effectiveness under asynchronous planning and imperfect communication by comparing our extension to the base version. Our extension does not result in any collisions or synchronization-induced deadlocks to which the base version is prone.

This paper addresses the problem of copying an unknown assembly of primitives with known shape and appearance using information extracted from a single photograph by an off-the-shelf procedure for object detection and pose estimation. The proposed algorithm uses a simple combination of physical stability constraints, convex optimization and Monte Carlo tree search to plan assemblies as sequences of pick-and-place operations represented by STRIPS operators. It is efficient and, most importantly, robust to the errors in object detection and pose estimation unavoidable in any real robotic system. The proposed approach is demonstrated with thorough experiments on a UR5 manipulator.