ICRA 2021 Papers

Predicting the future occupancy state of an environment is important to enable informed decisions for autonomous vehicles. Common challenges in occupancy prediction include vanishing dynamic objects and blurred predictions, especially for long prediction horizons. In this work, we propose a double-prong neural network architecture to predict the spatiotemporal evolution of the occupancy state. One prong is dedicated to predicting how the static environment will be observed by the moving ego vehicle. The other prong predicts how the dynamic objects in the environment will move. Experiments conducted on the real-world Waymo Open Dataset indicate that the fused output of the two prongs is capable of retaining dynamic objects and reducing blurriness in the predictions for longer time horizons than baseline models.

We are interested in learning models of non-stationary environments, which can be framed as a multitask learning problem. Model-free reinforcement learning algorithms can achieve good asymptotic performance in multitask learning at a cost of extensive sampling, due to their approach, which requires learning from scratch. While model-based approaches are among the most data efficient learning algorithms, they still struggle with complex tasks and model uncertainties. Meta-reinforcement learning addresses the efficiency and generalization challenges on multi task learning by quickly leveraging the meta-prior policy for a new task. In this paper, we propose a meta-reinforcement learning approach to learn the dynamic model of a non-stationary environment to be used for meta-policy optimization later. Due to the sample efficiency of model-based learning methods, we are able to simultaneously train both the meta-model of the non-stationary environment and the meta-policy until dynamic model convergence. Then, the meta-learned dynamic model of the environment will generate simulated data for meta-policy optimization. Our experiment demonstrates that our proposed method can meta-learn the policy in a non-stationary environment with the data efficiency of model-based learning approaches while achieving the high asymptotic performance of model-free meta-reinforcement learning.

Real-world missions require robots to detect objects in complex and changing environments. While deep learning methods for object detection are able to achieve a high level of performance, they can be unreliable when operating in environments that deviate from training conditions. However, by applying novelty detection techniques, we aim to build an architecture aware of when it cannot make reliable classifications, as well as identifying novel features/data. In this work, we have proposed and evaluated a system that assesses the competence of trained Convolutional Neural Networks (CNNs). This is achieved using three complementary introspection methods: (1) a Convolutional Variational Auto-Encoder (VAE), (2) a latent space Density-adjusted Distance Measure (DDM), and (3) a Spearman’s Rank Correlation (SRC) based approach. Finally these approaches are combined through a weighted sum, with weightings derived by maximising the correct attribution of novelty in an adversarial ‘meta-game’. Our experiments were conducted on real-world data from three datasets spread across two different domains: a planetary and an industrial setting. Results show that the proposed introspection methods are able to detect misclassifications and unknown classes indicative of novel features/data in both domains with up to 67% precision. Meanwhile classification results were either maintained or improved as a result.

We describe a docking mechanism and strategy to allow modular self-assembly for the Modboat: an inexpensive, underactuated, oscillating, surface-swimming robot powered by a single motor. Because propulsion is achieved through oscillation, orientation can be controlled only in the average; this complicates docking, which requires precise position and orientation control. Given these challenges, we present a docking strategy and a motion primitive for controlling orientation, and show that this strategy allows successful docking in multiple configurations. Moreover, we demonstrate that the Modboat is also capable of undocking and changing its dock configuration, all without any additional actuation. This is unique among similar modular robotic systems.

Aerial cinematography is significantly expanding the capabilities of film-makers. Recent progress in autonomous unmanned aerial vehicles (UAVs) has further increased the potential impact of aerial cameras, with systems that can safely track actors in unstructured cluttered environments. Professional productions, however, require the use of multiple cameras simultaneously to record different viewpoints of the same scene, which are edited into the final footage either in real time or in post-production. Such extreme motion coordination is particularly hard for unscripted action scenes, which are a common use case of aerial cameras. In this work we develop a real-time multi-UAV coordination system that is capable of recording dynamic targets while maximizing shot diversity and avoiding collisions and mutual visibility between cameras. We validate our approach in multiple cluttered environments of a photo-realistic simulator, and deploy the system using two UAVs in real-world experiments. We show that our coordination scheme has low computational cost and takes only 1.17 ms on average to plan for a team of 3 UAVs over a 10 s time horizon. Supplementary video: https://youtu.be/m2R3anv2ADE

It is essential in many applications to impose a scalable coordinated motion control on a large group of mobile robots, which is efficient in tasks requiring repetitive execution, such as environmental monitoring. In this paper, we design a guiding vector field to guide multiple robots to follow possibly different desired paths while coordinating their motions. The vector field uses a path parameter as a virtual coordinate that is communicated among neighboring robots. Then, the virtual coordinate is utilized to control the relative parametric displacement between robots along the paths. This enables us to design a saturated control algorithm for a Dubins-car-like model. The algorithm is distributed, scalable, and applicable for any smooth paths in an n-dimensional configuration space, and global convergence is guaranteed. Simulations with up to fifty robots and outdoor experiments with fixed-wing aircraft validate the theoretical results.

Visual-Inertial Odometry (VIO) has been widely used and researched to control and aid the automation of navigation of robots especially in the absence of absolute position measurements, such as GPS. However, when the observable landmarks in the scene lie far away, as in high-altitude flights for example, the fidelity of the metric scale estimate in VIO greatly degrades. Aiming to tackle this issue, in this work, we utilize the virtual stereo setup formed by two Unmanned Aerial Vehicles (UAVs), equipped with one camera and one Inertial Measurement Unit (IMU) each, exploiting their view overlap and relative distance measurements between them using onboard Ultra-Wideband (UWB) modules to enable collaborative VIO. In particular, we propose a decentralized collaborative estimation scheme, where each agent holds its own local map, achieving a low pose estimation latency, while ensuring consistency of each agents’ estimates via consensus-based optimization. Following a thorough evaluation in photorealistic simulations, we demonstrate the effectiveness of the approach at high-altitude flights of up to 160m, going significantly beyond the capabilities of state-of-the-art VIO methods. Finally, we show the advantage of actively adjusting the baseline on-the-fly over a fixed, target baseline, resulting in a significant reduction of the estimation error.

Multi-robot systems can perform task-related collaborative behaviors while maintaining connectivity within the system. However, some robots may fail to execute tasks or converge relatively slowly due to connectivity constraints. We consider the case that some robots may not have tasks assigned at a certain time frame, and they may help the task robots to achieve their goals by forming a connectivity graph with flexible topology. Therefore, we introduce a topology correction controller to provide flexibility for the task robots to perform task behaviors by modifying the topology of the connectivity graph for a faster convergence rate. We propose a distributed approach of blending weighted rendezvous and weighted flocking to form the correction controller. We prove that this scheme can guarantee a faster convergence rate and provide flexible connectivity graph topology. We then present our result of a system of up to thirty robots in various cluttered environments and show that our approach of behavior combination is robust and scalable.

In this paper, the distributed rendezvous control problem of networked uncertain robotic systems with bearing measurements is investigated. The network topology of the multi-robot systems is described by an undirected graph. The dynamics of robots is modeled by Euler-Lagrange equation with unknown inertial parameters, which is more general than simple kinematics considered in existing works on rendezvous problem of multi-robot systems. To achieve rendezvous, a distributed adaptive force/torque control law is developed for each robot, which uses bearings with respect to its neighbors instead of relative displacements or distances. It is shown that the resulting closed-loop multi-robot systems are globally asymptotically stable. Then, the rendezvous control problem of multiple wheeled mobile robots is further solved by the proposed approach. Finally, on-site experiment on networked TurtleBot3 Burger mobile robots is conducted and the results demonstrate effectiveness of the proposed approach.

Most recent works on multi-target tracking with multiple cameras focus on centralized systems. In contrast, this paper presents a multi-target tracking approach implemented in a distributed camera network. The advantages of distributed systems lie in lighter communication management, greater robustness to failures and local decision making. On the other hand, data association and information fusion are more challenging than in a centralized setup, mostly due to the lack of global and complete information. The proposed algorithm boosts the benefits of the Distributed-Consensus Kalman Filter with the support of a re-identification network and a distributed tracker manager module to facilitate consistent information. These techniques complement each other and facilitate the cross-camera data association in a simple and effective manner. We evaluate the whole system with known public data sets under different conditions demonstrating the advantages of combining all the modules. In addition, we compare our algorithm to some existing centralized tracking methods, outperforming their behavior in terms of accuracy and bandwidth usage.

This paper introduces the normalized unused sensing capacity to measure the amount of information that a sensor is currently gathering relative to its theoretical maximum. This quantity can be computed using entirely local information and works for arbitrary sensor models, unlike previous literature on the subject. This is then used to develop a distributed coverage control strategy for a team of heterogeneous sensors that automatically balances the load based on the current unused capacity of each team member. This algorithm is validated in a multi-target tracking scenario, yielding superior results to standard approaches that do not account for heterogeneity or current usage rates.

The implementation of optimization-based motion coordination approaches in real world multi-agent systems remains challenging due to their high computational complexity and potential deadlocks. This paper presents a distributed model predictive control (MPC) approach based on convex feasible set (CFS) algorithm for multi-vehicle motion coordination in autonomous driving. By using CFS to convexify the collision avoidance constraints, collision-free trajectories can be computed in real time. We analyze the potential deadlocks and show that a deadlock can be resolved by changing vehicles’ desired speeds. The MPC structure ensures that our algorithm is robust to low-level tracking errors. The proposed distributed method has been tested in multiple challenging multi-vehicle environments, including unstructured road, intersection, crossing, platoon formation, merging, and overtaking scenarios. The numerical results and comparison with other approaches (including a centralized MPC and reciprocal velocity obstacles) show that the proposed method is computationally efficient and robust, and avoids deadlocks.

Multi-Agent Path Finding (MAPF) is essential to large-scale robotic systems. Recent methods have applied reinforcement learning (RL) to learn decentralized polices in partially observable environments. A fundamental challenge of obtaining collision-free policy is that agents need to learn co-operation to handle congested situations. This paper combines communication with deep Q-learning to provide a novel learning based method for MAPF, where agents achieve cooperation via graph convolution. To guide RL algorithm on long-horizon goal-oriented tasks, we embed the potential choices of shortest paths from single source as heuristic guidance instead of using a specific path as in most existing works. Our method treats each agent independently and trains the model from a single agent’s perspective. The final trained policy is applied to each agent for decentralized execution. The whole system is distributed during training and is trained under a curriculum learning strategy. Empirical evaluation in obstacle-rich environment indicates the high success rate with low average step of our method.

In this paper we solve the full-consensus problem for multiple nonholonomic vehicles interacting over a directed leader-follower topology and subject to sensing constraints in the form of limited range and limited field-of-view. Remarkably, based on a polar-coordinates model transformation, the designed controller is time-invariant and smooth (in the domain of definition). Moreover, the control laws rely only on local measurements, making it well suited for implementation. The asymptotic convergence to the consensus manifold as well as the respect of the constraints is established using Lyapunov’s first method and cascaded systems theory. Realistic simulations in the Gazebo-ROS environment, which illustrate the effectiveness of our theoretical contributions, are shown in an accompanying video.

The technology of dynamic map fusion among networked vehicles has been developed to enlarge sensing ranges and improve sensing accuracies for individual vehicles. This paper proposes a federated learning (FL) based dynamic map fusion framework to achieve high map quality despite unknown numbers of objects in fields of view (FoVs), various sensing and model uncertainties, and missing data labels for online learning. The novelty of this work is threefold: (1) developing a three-stage fusion scheme to predict the number of objects effectively and to fuse multiple local maps with fidelity scores; (2) developing an FL algorithm which fine-tunes feature models (i.e., representation learning networks for feature extraction) distributively by aggregating model parameters; (3) developing a knowledge distillation method to generate FL training labels when data labels are unavailable. The proposed framework is implemented in the CARLA simulation platform. Extensive experimental results are provided to verify the superior performance and robustness of the developed map fusion and FL schemes.

Simultaneous localization and mapping (SLAM) is a crucial functionality for exploration robots and virtual/augmented reality (VR/AR) devices. However, some of such devices with limited resources cannot afford the computational or memory cost to run full SLAM algorithms. We propose a general client-server SLAM optimization framework that achieves accurate real-time state estimation on the device with low requirements of on-board resources. The resource-limited device (the client) only works on a small part of the map, and the rest of the map is processed by the server. By sending the summarized information of the rest of map to the client, the on-device state estimation is more accurate. Further improvement of accuracy is achieved in the presence of on-device early loop closures, which enables reloading useful variables from the server to the client. Experimental results from both synthetic and real-world datasets demonstrate that the proposed optimization framework achieves accurate estimation in real-time with limited computation and memory budget of the device.

We present a hierarchical planning and control framework that enables an agent to perform various tasks and adapt to a new task flexibly. Rather than learning an individual policy for each particular task, the proposed framework, DISH, distills a hierarchical policy from a set of tasks by representation and reinforcement learning. The framework is based on the idea of latent variable models that represent high-dimensional observations using low-dimensional latent variables. The resulting policy consists of two levels of hierarchy: (i) a planning module that reasons a sequence of latent intentions that would lead to an optimistic future and (ii) a feedback control policy, shared across the tasks, that executes the inferred intention. Because the planning is performed in low-dimensional latent space, the learned policy can immediately be used to solve or adapt to new tasks without additional training. We demonstrate the proposed framework can learn compact representations (3- and 1-dimensional latent states and commands for a humanoid with 197- and 36-dimensional state features and actions) while solving a small number of imitation tasks, and the resulting policy is directly applicable to other types of tasks, i.e., navigation in cluttered environments.

Search-based planning with motion primitives is a powerful motion planning technique that can provide dynamic feasibility, optimality, and real-time computation times on size, weight, and power-constrained platforms in unstructured environments. However, optimal design of the motion planning graph, while crucial to the performance of the planner, has not been a main focus of prior work. This paper proposes to address this by introducing a method of choosing vertices and edges in a motion primitive graph that is grounded in sampling theory and leads to theoretical guarantees on planner completeness. By minimizing dispersion of the graph vertices in the metric space induced by trajectory cost, we optimally cover the space of feasible trajectories with our motion primitive graph. In comparison with baseline motion primitives defined by uniform input space sampling, our motion primitive graphs have lower dispersion, find a plan with fewer iterations of the graph search, and have only one parameter to tune.

Semantic segmentation of surgical instruments provides essential priors for autonomous surgery. This task is however challenging since the fine-structure of surgical instruments requires the accurate segmentation of detailed regions in images. As the visual guidance for autonomous surgery, the algorithm should also be real-time and friendly to embedded systems. In this paper, a discriminative asymmetric learning framework is proposed to balance the efficiency and effectiveness of surgical instrument segmentation. Two convolutional neural networks with specific designs are deployed to extract the detail and semantic features of instruments. To reduce the redundancy of visual representation, the aggregator-discriminator mechanism is proposed to distinguish the features learned from different levels. Experiments demonstrate that the proposed method contributes to competitive segmentation accuracy and a higher efficiency compared to existing methods.