In this paper, a global descriptor for a LiDAR point cloud, called LiDAR Iris, is proposed for fast and accurate loop-closure detection. A binary signature image can be obtained for each point cloud after several LoG-Gabor filtering and thresholding operations on the LiDAR-Iris image representation. Given two point clouds, their similarities can be calculated as the Hamming distance of two corresponding binary signature images extracted from the two point clouds, respectively. Our LiDAR-Iris method can achieve a pose-invariant loop-closure detection at a descriptor level with the Fourier transform of the LiDAR-Iris representation if assuming a 3D (x,y,yaw) pose space, although our method can generally be applied to a 6D pose space by re-aligning point clouds with an additional IMU sensor. Experimental results on five road-scene sequences demonstrate its excellent performance in loop-closure detection.

The ability of legged systems to traverse highly-constrained environments depends by and large on the performance of their motion and balance controllers. This paper presents a controller that excels in a scenario that most state-of-the-art balance controllers have not yet addressed: line walking, or walking on nearly null support regions. Our approach uses a low-dimensional virtual model (2-DoF) to generate balancing actions through a previously derived four-term balance controller and transforms them to the robot through a derived kinematic mapping. The capabilities of this controller are tested in simulation, where we show the 90kg quadruped robot HyQ crossing a bridge of only 6 cm width (compared to its 4 cm diameter foot sphere), by balancing on two feet at any time while moving along a line. Additional simulations are carried to test the performance of the controller and the effect of external disturbances. The same controller is then used on the real robot to present for the first time a legged robot balancing on a contact line of nearly null support area.

This report studies the emergent behavior of systems of agents performing cyclic pursuit controlled by an external broadcast signal detected by a random set of the agents. Two types of cyclic pursuit are analyzed: 1)linear cyclic pursuit, where each agent senses the relative position of its target or leading agent 2)non-linear cyclic pursuit, where the agents can sense only bearing to their leading agent and colliding agents merge and continue on the path of the pursued agent (a so-called "bugs" model). Cyclic pursuit is, in both cases, a gathering algorithm, which has been previously analyzed. The novelty of our work is the derivation of emergent behaviours, in both linear and non-linear cyclic pursuit, in the presence of an exogenous broadcast control detected by a random subset of agents. We show that the emergent behavior of the swarm depends on the type of cyclic pursuit. In the linear case, the agents asymptotically align in the desired direction and move with a common speed which is a proportional to the ratio of the number of agents detecting the broadcast control to the total number of agents in the swarm, for any magnitude of input (velocity) signal. In the non-linear case, the agents gather and move with a shared velocity, which equals the input velocity signal, independently of the number of agents detecting the broadcast signal.

Automotive cameras, particularly surround-view cameras, tend to get soiled by mud, water, snow, etc. For higher levels of autonomous driving, it is necessary to have a soiling detection algorithm which will trigger an automatic cleaning system. Localized detection of soiling in an image is necessary to control the cleaning system. It is also necessary to enable partial functionality in unsoiled areas while reducing confidence in soiled areas. Although this can be solved using a semantic segmentation task, we explore a more efficient solution targeting deployment in low power embedded system. We propose a novel method to regress the area of each soiling type within a tile directly. We refer to this as coverage. The proposed approach is better than learning the dominant class in a tile as multiple soiling types occur within a tile commonly. It also has the advantage of dealing with coarse polygon annotation, which will cause the segmentation task. The proposed soiling coverage decoder is an order of magnitude faster than an equivalent segmentation decoder. We also integrated it into an object detection and semantic segmentation multi-task model using an asynchronous back-propagation algorithm. A portion of the dataset used will be released publicly as part of our WoodScape dataset to encourage further research.

We present a reinforcement learning based framework for human-centered collaborative systems. The framework is proactive and balances the benefits of timely actions with the risk of taking improper actions by minimizing the total time spent to complete the task. The framework is learned end-to-end in an unsupervised fashion addressing the perception uncertainties and decision making in an integrated manner. The framework is shown to provide more fluent coordination between human and robot partners on an example task of packaging compared to alternatives for which perception and decision-making systems are learned independently, using supervised learning. The foremost benefit of the proposed approach is that it allows for fast adaptation to new human partners and tasks since tedious annotation of motion data is avoided and the learning is performed on-line.

Many kinds of lower-limb exoskeletons were developed for walking assistance. However, time-delay arised from the computation time and the communication delays is still a general problem when controlling these exoskeletons. In this research, we proposed a novel method to prevent the time-delay when controlling a walking assist exoskeleton by predicting future plantar force and walking status. By using Long Short-term Memory (LSTM) and fully-connected network, the plantar force can be predicted only from the data measured from inertial measurement unit (IMU) sensors, not only during the periodic walking but also when the start and end of walking. The walking status and the desired assisting timing can also be detected from the predicted plantar force. By sending the control command beforehand with considering the time-delay, the exoskeleton can be moved right on the desired assisting timing. In the experiments, the prediction errors of the plantar force and the assisting timing are confirmed. The performance of the proposed method is also evaluated by controlling the exoskeleton using the trained model.

Recent huge technological development of Unmanned Aerial Vehicles (UAVs) can provide breakthrough means of fighting wildland fires. We propose an innovative forest firefighting system based on the use of a swarm of hundreds of UAVs able to generate a continuous flow of extinguishing liquid on the fire front, simulating the rain effect. Automatic battery replacement and refilling of the extinguishing liquid ensure the continuity of the action. We demonstrate the validity of the approach first computing the critical water flow rate according to the main factors involved in the evolution of a fire, then estimating the number of linear meters of active fire front that can be extinguished depending on the number of drones available and the amount of extinguishing fluid carried. A fire propagation cellular automata model is also employed to study the evolution of the fire. The results suggest that the proposed system can successfully integrate, or in some cases completely replace, current forest firefighting techniques.

We present the design of a new robotic human augmentation system that will assist the operator in carrying a heavy payload, reaching and maintaining difficult postures, and ultimately better performing their job. The Extra Robotic Legs (XRL) system is worn by the operator and consists of two articulated robotic legs that move with the operator to bear a heavy payload. The design was driven by a need to increase the effectiveness of hazardous material emergency response personnel who are encumbered by their personal protective equipment (PPE). The legs will ultimately walk, climb stairs, crouch down, and crawl with the operator while eliminating all external PPE loads on the operator. The forces involved in the most extreme loading cases were analyzed to find an effective strategy for reducing actuator loads. The analysis reveals that the maximum torque is exerted during the transition from the crawling to standing mode of motion. Peak torques are significantly reduced by leveraging redundancy in force application resulting from a closed-loop kinematic chain formed by a particular posture of the XRL. The actuators, power systems, and transmission elements were designed from the results of these analyses. Using differential mechanisms to combine the inputs of multiple actuators into a single degree of freedom, the gear reductions needed to bear the heavy loads could be kept at a minimum, enabling high bandwidth force control due to the near-direct-drive transmission. A prototype was fabricated utilizing the insights gained from these analyses and initial tests indicate the feasibility of the XRL system.

Deep reinforcement learning (DRL) has been proven to be a powerful paradigm for learning complex control policy autonomously. Numerous recent applications of DRL in robotic grasping have successfully trained DRL robotic agents end-to-end, mapping visual inputs into control instructions directly, but the amount of training data required may hinder these applications in practice. In this paper, we propose a DRL based robotic visual grasping framework, in which visual perception and control policy are trained separately rather than end-to-end. The visual perception produces physical descriptions of grasped objects and the policy takes use of them to decide optimal actions based on DRL. Benefiting from the explicit representation of objects, the policy is expected to be endowed with more generalization power over new objects and environments. In addition, the policy can be trained in simulation and transferred in real robotic system without any further training. We evaluate our framework in a real world robotic system on a number of robotic grasping tasks, such as semantic grasping, clustered object grasping, moving object grasping. The results show impressive robustness and generalization of our system.

To perform tasks well in a new domain, one must first know something about it. This paper reports on a robot controller for navigation through unfamiliar indoor worlds. Based on spatial affordances, it integrates planning with reactive heuristics. Before it addresses specific targets, however, the system deliberately explores for high-level connectivity and captures that data in a cognitive spatial model. Despite limited exploration time, planning in the resultant model is faster and better supports successful travel in a challenging, realistic space.

High dimensional black-box optimization has broad applications but remains a challenging problem to solve. Given a set of samples $\{\vx_i, y_i\}$, building a global model (like Bayesian Optimization (BO)) suffers from the curse of dimensionality in the high-dimensional search space, while a greedy search may lead to sub-optimality. By recursively splitting the search space into regions with high/low function values, recent works like LaNAS shows good performance in Neural Architecture Search (NAS), reducing the sample complexity empirically. In this paper, we coin LA-MCTS that extends LaNAS to other domains. Unlike previous approaches, LA-MCTS learns the partition of the search space using a few samples and their function values in an online fashion. While LaNAS uses linear partition and performs uniform sampling in each region, our LA-MCTS adopts a nonlinear decision boundary and learns a local model to pick good candidates. If the nonlinear partition function and the local model fits well with ground-truth black-box function, then good partitions and candidates can be reached with much fewer samples. LA-MCTS serves as a \emph{meta-algorithm} by using existing black-box optimizers (e.g., BO, TuRBO) as its local models, achieving strong performance in general black-box optimization and reinforcement learning benchmarks, in particular for high-dimensional problems.