IROS 2018 Papers

This paper presents the dynamic modelling and motion planning method for a quadruped robot that uses its legs for nonprehensile manipulation as well as locomotion. Three different working modes named Drive Mode, Inchworm Mode and Scoot Mode are proposed to enable the robot to move forward together with the object. We firstly introduce a universal model for these modes and deduce its dynamic equation. Then the contact force constraints are combined and mapped to the system state variables. Based on the acquired state acceleration constraints, the motion planning problem can be solved by designing system state paths in the phase space. After that, we described the mathematical problems within the three working modes and generate the robot motions accordingly. Finally, experimental results obtained through simulations and physical tests are reported to demonstrate the effectiveness of our method.

One of the most concerning and costly problems in hospitals is patients falls. We address this problem by introducing PAM, a patient assistant mobile robot, that maneuvers mobility aids to assist with fall prevention. Common objects found inside hospitals include objects with legs (i.e. walkers, tables, chairs, equipment stands). For a mobile robot operating in such environments, safely maneuvering these objects without collision is essential. Since providing the robot with dynamic models of all possible legged objects that may exist in such environments is not feasible, autonomous learning of an approximate dynamic model for these objects would significantly improve manipulation planning. We describe a probabilistic method to do this by fitting pre-categorized object models learned from minimal force and motion interactions with an object. In addition, we account for multiple manipulation strategies, which requires a hybrid control system comprised of discrete grasps on legs and continuous applied forces. To do this, we use a simple one-wheel point-mass model. A hybrid MPC-based manipulation planning algorithm was developed to compensate for modeling errors. While the proposed algorithm applies to a broad range of legged objects, we only show results for the case of a 2-wheel, 4-legged walker in this paper. Simulation and experimental tests show that the obtained dynamic model is sufficiently accurate for safe and collision-free manipulation. When combined with the proposed manipulation planning algorithm, the robot can successfully move the object to a desired position without collision.

This paper presents an implementation of model predictive control (MPC) to determine ground reaction forces for a torque-controlled quadruped robot. The robot dynamics are simplified to formulate the problem as convex optimization while still capturing the full 3D nature of the system. With the simplified model, ground reaction force planning problems are formulated for prediction horizons of up to 0.5 seconds, and are solved to optimality in under 1 ms at a rate of 20-30 Hz. Despite using a simplified model, the robot is capable of robust locomotion at a variety of speeds. Experimental results demonstrate control of gaits including stand, trot, flying-trot, pronk, bound, pace, a 3-legged gait, and a full 3D gallop. The robot achieved forward speeds of up to 3 m/s, lateral speeds up to 1 m/s, and angular speeds up to 180 deg/sec. Our approach is general enough to perform all these behaviors with the same set of gains and weights.

In this paper, Dynamic Dumbbell, a novel robotic device for advanced muscular exercise of upper limb is presented. The type of exercise load is classified and designed in terms of mechanical engineering to be implemented in Dynamic Dumbbell. The exercise load model, which is named as programmable exercise load, is realized by Dynamic Dumbbell. To generate the programmable exercise load, two of compact Planetary-geared Elastic Actuator, which is a rotary Series Elastic Actuator (SEA), are utilized in Dynamic Dumbbell. The SEAs are controlled using high performance force control algorithm. Experimental results verifies the effectiveness of the proposed Dynamic Dumbbell and programmable exercise load.

Deforming a cable to a desired (reachable) shape is a trivial task for a human to do without even knowing the internal dynamics of the cable. This paper proposes a framework for cable shapes manipulation with multiple robot manipulators. The shape is parameterized by a Fourier series. A local deformation model of the cable is estimated on-line with the shape parameters. Using the deformation model, a velocity control law is applied on the robot to deform the cable into the desired shape. Experiments on a dual-arm manipulator are conducted to validate the framework.

To make production lines more flexible, dual-arm robots are good candidates to be deployed in autonomous assembly units. In this paper, we propose a sparse kinematic control strategy, that minimizes the number of joints actuated for a coordinated task between two arms. The control strategy is based on a hierarchical sparse QP architecture. We present experimental results that highlight the capability of this architecture to produce sparser motions (for an assembly task) than those obtained with standard controllers.

Capturing a moving object with large momentum by a dual-arm robot is especially challenging because of the requirement of dual-arm coordinated motion planning for tracking the moving object, and the operational force control for contact and momentum transfer. In this paper, we present a dual-arm coordinated motion planning and compliance control method with a unique null-space projected relative Jacobian and relative operational force between the two arms. The proposed method is able to plan dual-arm capturing motion and control the capturing force without disturbing the tracking motion. We have also adopted a direct collocation trajectory optimization method to generate optimal trajectory to decrease the object's momentum with minimum effort. Simulation and experiment of dual-arm robots picking up a moving box on a mobile platform are carried out to verify the proposed method.

Obstacle avoidance is a key feature for safe Unmanned Aerial Vehicle (UAV) navigation. While solutions have been proposed for static obstacle avoidance, systems enabling avoidance of dynamic objects, such as drones, are hard to implement due to the detection range and field-of-view (FOV) requirements, as well as the constraints for integrating such systems on-board small UAVs. In this work, a dataset of 6k synthetic depth maps of drones has been generated and used to train a state-of-the-art deep learning-based drone detection model. While many sensing technologies can only provide relative altitude and azimuth of an obstacle, our depth map-based approach enables full 3D localization of the obstacle. This is extremely useful for collision avoidance, as 3D localization of detected drones is key to perform efficient collision-free path planning. The proposed detection technique has been validated in several real depth map sequences, with multiple types of drones flying at up to 2 m/s, achieving an average precision of 98.7 %, an average recall of 74.7 % and a record detection range of 9.5 meters.

Machines with high levels of autonomy such as robots and our growing need to interact with them creates challenges to ensure safe operation. The recent interest to create autonomous vehicles through the integration of control and decision-making systems makes such vehicles robots too. We therefore applied estimation and decision-making mechanisms currently investigated for human-robot interaction to human-vehicle interaction. In other words, we define the vehicle as an autonomous agent with which the human driver interacts, and focus on understanding the human intentions and decision-making processes. These are then integrated into the ro-bot`s/vehicle's own control and decision-making system not only to understand human behaviour while it occurs but to predict the next actions. To obtain knowledge about the human's intentions, this work relies heavily on the use of motion tracking data (i.e. skeletal tracking, body posture)gathered from drivers whilst driving. We use a data-driven approach to both classify current driving manoeuvres and predict future manoeuvres, by using a fixed prediction window and augmenting a standard set of manoeuvres. Results are validated against drivers of different sizes, seat preferences and levels of driving expertise to evaluate the robustness of the methods; precision and recall metrics higher than 95% for manoeuvre classification and 90% for manoeuvre prediction with time-windows of up to 1.3 seconds are obtained. The idea of prediction adds a highly novel aspect to human-robot/human-vehicle interaction, allowing for decision and control at a later point.

In this paper we address the problem of flow-aware trajectory planning in dynamic environments considering flow model uncertainty. Flow-aware planning aims to plan trajectories that adhere to existing flow motion patterns in the environment, with the goal to make robots more efficient, less intrusive and safer. We use a statistical model called CLiFF-map that can map flow patterns for both continuous media and discrete objects. We propose novel cost and biasing functions for an RRT* planning algorithm, which exploits all the information available in the CLiFF-map model, including uncertainties due to flow variability or partial observability. Qualitatively, a benefit of our approach is that it can also be tuned to yield trajectories with different qualities such as exploratory or cautious, depending on application requirements. Quantitatively, we demonstrate that our approach produces more flow-compliant trajectories, compared to two baselines.

Deep learning-based robotic grasping has made significant progress thanks to algorithmic improvements and increased data availability. However, state-of-the-art models are often trained on as few as hundreds or thousands of unique object instances, and as a result generalization can be a challenge. In this work, we explore a novel data generation pipeline for training a deep neural network to perform grasp planning that applies the idea of domain randomization to object synthesis. We generate millions of unique, unrealistic procedurally generated objects, and train a deep neural network to perform grasp planning on these objects. Since the distribution of successful grasps for a given object can be highly multimodal, we propose an autoregressive grasp planning model that maps sensor inputs of a scene to a probability distribution over possible grasps. This model allows us to sample grasps efficiently at test time (or avoid sampling entirely). We evaluate our model architecture and data generation pipeline in simulation and the real world. We find we can achieve a >90% success rate on previously unseen realistic objects at test time in simulation despite having only been trained on random objects. We also demonstrate an 80% success rate on real-world grasp attempts despite having only been trained on random simulated objects.

We present Dolphin, an extensible programming language for autonomous vehicle networks. A Dolphin program expresses an orchestrated execution of tasks defined compositionally for multiple vehicles. Building upon the base case of elementary one-vehicle tasks, the built-in operators include support for composing tasks in several forms, for instance according to concurrent, sequential, or event-based task flow. The language is implemented as a Groovy DSL, facilitating extension and integration with external software packages, in particular robotic toolkits. The paper describes the Dolphin language, its integration with an open-source toolchain for autonomous vehicles, and results from field tests using unmanned underwater vehicles (UUVs) and unmanned aerial vehicles (UAVs).

Culture, amongst other individual and social factors, plays a crucial role in human-human interactions. If robots should become a part of our society, they should be able to act in culture-specific manners as well. In this paper, we showcase the implementation of a cultural dichotomy, namely individualism vs. collectivism, in a social robots' conversation. Presenting these conversations to human observers from Germany and Japan, we investigate whether the implemented differences are recognized as such, and whether stereotypical culture-specific behaviours that correspond to the observers' cultural background is preferred. Results suggest that the manipulations in behaviour had the intended effect, but are not reflected in personal preferences.

The motivation of the work presented here is to create a component of a safety system based on 3D lidar sensors, specifically for industrial environments where some rules can be set for people who will be in close proximity to working robots. Specifically, the operating procedure that is put in place in the workplace is that all people must wear the provided high visibility clothing, which has retro-reflective strips attached. It is shown here that the retro-reflective strips provide a strong cue for pedestrian detection in the intensity data from a lidar sensor within a range of 4 metres. We present and compare multiple methods of exploiting this cue and provide a recommendation for how a safety system should be architected in order to best exploit the lidar intensity data in combination with more common approaches for detection of objects from the lidar range data. Amongst these detection methods is the use of neural networks, which present challenges for key components of standardized safety system development-in particular, for programming methodology control, interpretability of testing and diagnostic coverage. We propose methods for how to start to address these challenges and how to integrate neural networks into safety systems.

We propose a distributionally robust incremental sampling-based method for kinodynamic motion planning under uncertainty, which we call distributionally robust RRT (DR-RRT). In contrast to many approaches that assume Gaussian distributions for uncertain parameters, here we consider moment-based ambiguity sets of distributions with given mean and covariance. Chance constraints for obstacle avoidance and internal state bounds are then enforced under the worst-case distribution in the ambiguity set, which gives a coherent assessment of constraint violation risks. The method generates risk-bounded trajectories and feedback control laws for robots operating in dynamic, cluttered, and uncertain environments, explicitly incorporating localization error, stochastic process disturbances, unpredictable obstacle motion, and uncertain obstacle location. We show that the algorithm is probabilistically complete under mild assumptions. Numerical experiments illustrate the effectiveness of the algorithm.

This paper considers the combination of temporal logic (TL) specifications and local objective functions to create online, multiagent, motion plans. These plans are guaranteed to satisfy a persistent mission TL specification and locally optimize an objective function (e.g. in this paper, a cost based on information entropy). The presented approach decouples the two tasks by assigning sub-teams of agents to fulfill the TL specification, while unassigned agents optimize the objective function locally. This paper also presents a novel decoupling of the classic product automaton based approach while maintaining satisfaction guarantees. We also qualitatively show that optimality loss in the local greedy minimization due to the TL constraints can be approximated based on specification complexity. This approach is evaluated with a set of simulations and an experiment of 6 robots with real sensors.

This video presents a distributed trajectory generator for formation control of multi-robot systems. The desired formation is defined by its geometric parameters but the position of each robot in the formation is not predefined a priori. The contribution is the design of a distributed algorithm to compute the robots' positions with respect to a given target while maintaining a particular formation which can be reconfigured on-line. A tracking controller ensures the convergence of the robots to their desired positions.

Autonomous aerial transportation will be a fixture of future robotic societies, simultaneously requiring more stringent safety requirements and fewer resources for characterization than current commercial air transportation. More robust, adaptable, self-state estimation will be necessary to create such autonomous systems. We present a modular, scalable, distributed pressure sensing skin for aerodynamic state estimation of a large, flexible aerostructure. This skin used a network of 22 nodes that performed in situ computation and communication of data collected from 74 pressure sensors, which were embedded into the skin panels of an ultra-lightweight 14-foot wingspan made from commutable, lattice-based subcomponents, and tested at NASA Langley Research Center's 14X22 wind tunnel. The density of the pressure sensors allowed for the use of a novel distributed algorithm to generate estimates of the wing lift contribution that were more accurate than the direct integration of the pressure distribution over the wing surface.

This study proposes a Direction of Arrival (DoA)-aided attack detection scheme to identify cyberattacks on networked multi-robot systems. For each agent, a local estimator is designed to generate robust residuals, and a parametric statistical tool corresponding to the residuals is elaborated to build sensitive decision rules. These locally stored residuals and thresholds are shared between robots via a wireless network, allowing a multi-robot system to complete its mission in the presence of one or more compromised agents. The proposed DoA-aided attack detection scheme is tested on a multi-robot testbed with a team of 10 robots. Experimental results demonstrate that the proposed detection scheme enables each robot to identify malicious activities without shearing the global coordination.