Diffusion models have seen rapid adoption in robotic imitation learning, enabling autonomous execution of complex dexterous tasks. However, action synthesis is often slow, requiring many steps of iterative denoising, limiting the extent to which models can be used in tasks that require fast reactive policies. To sidestep this, recent works have explored how the distillation of the diffusion process can be used to accelerate policy synthesis. However, distillation is computationally expensive and can hurt both the accuracy and diversity of synthesized actions. We propose SDP (Streaming Diffusion Policy), an alternative method to accelerate policy synthesis, leveraging the insight that generating a partially denoised action trajectory is substantially faster than a full output action trajectory. At each observation, our approach outputs a partially denoised action trajectory with variable levels of noise corruption, where the immediate action to execute is noise-free, with subsequent actions having increasing levels of noise and uncertainty. The partially denoised action trajectory for a new observation can then be quickly generated by applying a few steps of denoising to the previously predicted noisy action trajectory (rolled over by one timestep). We illustrate the efficacy of this approach, dramatically speeding up policy synthesis while preserving performance across both simulated and real-world settings. Project website: https://streaming-diffusion-policy.github.io.

Automation and the use of robots for welding operations is an important research topic. Being able to automate and, thus, save time for setting up and using robotic welding for complex, large-scale structures made of reflective materials, such as aluminium, will provide clear economic and competitive advantages. However, challenges coming from the ability to accurately detect and calibrate the robot for a given physical workpiece in addition to noises, such as the reflections, make it hard to develop and demonstrate a feasible automation solution. This paper proposes combining laser line scanning technology with CAD-based analysis of a workpiece geometry to support the identification of relevant elements of the workpiece in the physical world and thus support welding operations. An extendable trigger definition method is proposed to identify features of interest in a workpiece. The method can potentially support the execution of welding sequences, which in our case can be represented as a sequence of triggers that have to be observed and followed at the robot runtime to weld the workpiece together.

In this article, we present a new procedure for the derivation of the linearized kinematics and dynamics of a flexible industrial robotic manipulator. We introduce the Lie groups of dual rotation and dual homogeneous transformation matrices, which are the basis for the derivation of dual twists. In addition, dual screws and dual screw transformation matrices are introduced, which are used for the development of a general and systematic linearization procedure for both kinematics and dynamics. This leads to expressions that are linearized in all the states associated with the elastic motion. The dynamic modeling procedure is based on Kane's method, where the partial velocities and partial angular velocities are given as dual screws arranged as columns of dual projection matrices. The elasticity of the links is modeled by the assumed mode method. The presented procedure is implemented numerically for a 4-degrees of freedom manipulator and the simulation results are given.

This paper presents a method for coarse alignment of point clouds by introducing a new descriptor based on the local curvature. The method is developed for model fitting a CAD model for use in robotic assembly. The method is based on selecting keypoints depending on shape factors calculated from the local covariance matrix of the surface. A descriptor is then calculated for each keypoint by fitting two spheres that describe the curvature of the surface. The spheres are calculated using conformal geometric algebra, which gives a convenient and efficient description of the geometry. The keypoint descriptors for the model and the observed point cloud are then compared to estimate the corresponding keypoints, which are used to calculate the displacement. The method is tested in several experiments. One experiment is for robotic assembly, where objects are placed on a table and their position and orientation is estimated using a 3-D CAD model. Note to Practitioners- 3-D cameras can be used in robotic assembly for recognizing objects, and for determining the position and orientation of parts to be assembled. In such applications, 3-D CAD models will be available for the objects, and point clouds representing each object can be generated for comparison with the observed point clouds from the 3-D camera. It is not straightforward to use existing descriptors in this paper, as the point cloud from the CAD model and the observed point cloud may differ due to different viewpoints and potential occlusions. The method proposed in this paper is intended to be easy to apply to industrial assembly problems where there is a need for a robust estimation of the displacement of an object, either as a coarse estimate for use in grasping or as an initial guess to use in fine registration for demanding assembly operations with close tolerances. The method exploits the curvature of the point clouds to accurately describe the surrounding surface of each point. This method serves as a basis for future industrial implementations.

In this paper we propose an efficient method for calculating the forces of constraints in an open-chain multibody system, like a robot mounted on a vehicle with 6 degrees of freedom. The dynamical model is based on Kane's equation of motion, where screw theory is used to calculate the projection matrices from the link twists of the multibody system. This leads to a general modeling procedure relying on screw transformations that is presented in the paper. The procedure for determination of the constraint forces is given as an extension of the dynamical model and can be implemented after the equations of motion have been formulated and solved. We implement the described method for the specific case of a vessel with a heavy crane, and provide the simulation results. The method provides a basis for future work on the detailed modeling of friction in the joints of serial link mechanisms, and on the evaluation of potential fatigue consequences of different control solutions.

This paper presents an approach for initial alignment or coarse registration of a partial 3D point cloud of objects. The method is based on computing the centroid of the points in the point cloud, and a line derived from the surface normals. This approach uses conformal geometric algebra and non-linear least squares optimization to achieve the results. The method performs well in experiments, and it is shown that it performs more accurately the more points are sampled.

This paper presents the development of a robotic cleaning solution for a fish processing plant. The project is currently at the stage of a first prototype consisting of a serial manipulator, a vertical linear axis and a rotational axis for the vertical axis. The purpose of the prototype is to validate the cleaning quality of a robotic cleaning solution. A cleaning solution will have to spray the equipment and machines in the processing plant with chemicals and water to remove fish residue and bacteria, and special design considerations have to be taken with regards to water proofing and corrosion resistance. In order to validate such a system, a cleaning test was performed on an electric stunner, a machine typically found in salmon slaughterhouses. Results from the cleaning test show that robotic cleaning of fish processing equipment gives promising results. However, several issues related to making a system that can clean a whole plant are not resolved. Further work will require the development of a custom serial manipulator and a custom linear axis for navigating the manipulator. The main purpose of this paper is to present design considerations and investigate the validity of a robotic cleaning solution aimed at fish processing plants. This research is at TRL 5, with validation of technology in relevant environment.

This paper presents a novel method for detecting collisions of a swinging crane load using a particle filter for visual tracking. The particle filter uses a dynamical model of the nominal motion of the crane load, and an alternative dynamical model that describes the crane load under collision. A jump Markov model is formulated with dynamics described by one of the models. The particle filter detects whether the object motion is due to the nominal model or the alternative model. The particle filter tracks markers placed on the crane load. The performance of the proposed method is demonstrated in laboratory experiments.

This paper studies control properties of the dynamics of underactuated vehicles (e.g. underactuated surface vessels, underwater vehicles, aeroplanes or spacecraft). The unactuated dynamics implies constraints on the accelerations. Both the necessary and sufficient conditions for these constraints to be second-order nonholonomic, first-order nonholonomic or holonomic are developed. It is shown that underactuated vehicles with a gravitational field where the elements corresponding to the unactuated dynamics are zero, are not C/sup 1/ asymptotically stabilizable to a single equilibrium.

Presents a feedback control law which gives exponential stabilization of a nonholonomic underwater vehicle to a constant desired configuration. This is achieved by transforming the kinematic model into a chained form, and applying a control scheme for exponential stabilization of chained form systems. The nonholonomic vehicle under consideration is described by a kinematic model with three angular velocities and one translational velocity component as inputs. The kinematic model of the underwater vehicle is given in SE(3) by a position vector and a rotation matrix. Due to the difference in topology, the transformation into chained form is not global, but has singularities for large angular deviations. This problem is discussed, and conditions for avoiding the singularities are derived.>

It is proved that if the manipulator Jacobian is properly scaled, the largest singular value is bounded within a small interval close to unity. In this case, the inverse of the smallest singular value is a good estimate of the condition number. This is useful in singularity analysis. Bounds are derived for a general six-joint manipulator and for the ABB IRb 2000 industrial robot in a case study.>

Energy-optimal trajectories for an autonomous underwater vehicle can be computed using a numerical solution of the optimal control problem. The vehicle is modeled with the six-dimensional nonlinear and coupled equations of motion, controlled with DC-motors in all degrees of freedom. The actuators are modeled and controlled with velocity loops. The dissipated energy is expressed in terms of the control variables as a nonquadratic function. Two different numerical methods are used in this study, a function space conjugate gradient (CG) method, and a control vector parameterization (CVP) method. Energy-optimal trajectories were computed in simulation experiments. Good results were achieved with the CVP method, which was superior to the CG method.>

Singularity robust redundancy resolution with task priority can be implemented using the extended Jacobian technique with weighted damped least-squares. The resulting scheme is simple to implement and involves less computation than the task priority scheme. The minimum singular value of the Jacobian can be estimated reliably and accurately with little computation, and this estimate was used to calculate an appropriate damping factor. A constant damping factor was also used with good results. The scheme was successfully implemented in a simulation study with a seven-joint manipulator with a kinematic design derived from the PUMA geometry.>

A control scheme for the coordination of motion in a macro-micro manipulator system is presented. The motion of the end-effector is decoupled from the rest of the system using resolved acceleration control, while a slow gross positioning is used for the macro part. To avoid saturated inputs and excessive use of energy for systems with limited fuel, the end-effector is not decoupled from macro part motion, but from selected directions of the linear momentum of the total system. For the vehicle-manipulator case, the end-effector is decoupled also from the angular momentum. This leads to lower force and torque peaks in the actuators, and control force is used on the macro part only when it is necessary to reposition this to keep the micro part inside its workspace.>

A solution to the problem of singularities for six-joint manipulators is presented. By using this method, the unachievable components of the commanded motion are removed, while an exact inverse kinematic solution is used for the remaining motion components. At a singularity there are directions in Cartesian space where a differential translation or rotation cannot be specified. The underlying area is to identify these directions and then to eliminate the corresponding components of the commanded motion when the manipulator becomes singular using a pseudoinverse of the manipulator Jacobian. In order to avoid excessive joint velocities close to the singularities, the manipulator is treated as singular in the neighborhood of the singularity. A continuous solution is achieved by interpolation in the degenerate directions.>

A control system for a 7-joint manipulator with three intersecting shoulder axes is presented. A singularity avoidance scheme is developed, where the redundancy is resolved through a position transformation. The internal and external motions are decoupled using nonlinear feedback linearization in an augmented task space. It is then possible to use linear controllers both for the internal nullspace and for the end-effector coordinates. It is shown how optimal control theory can be applied to coordinate the motion in the arm, and how to do this so that the input generalized forces are minimized. This gives a coordination of internal and external motion which resembles the motion of the human arm; this is demonstrated by simulating a throwing motion with the manipulator.>

The reference trajectory is normally known in advance in manipulator control, which makes it possible to apply linear quadratic optimal tracking. This gives a control system which rounds corners and generates optimal feedforward. The method may be used for references consisting of straight-line segments as an alternative to the two-step method of using splines to smooth the reference and then applying feedforward. in addition, the method can be used for more complex trajectories. The actual dynamics of the manipulator are taken into account, and this results in smooth and accurate tracking. The method has been applied in combination with the computed torque technique and excellent performance was demonstrated in a simulation study. The method has also been applied experimentally to an industrial spray-painting robot where a saw-tooth reference was tracked. The corner was rounded well, and the steady-state tracking error was eliminated by the optimal feedforward.>

A controller for redundant manipulators with a small, fast manipulator mounted on a positioning part has been developed. The controller distributes the fast motion to the small, fast manipulator and the slow, gross motion to the positioning part. A position reference is generated on-line to the positioning part to avoid singularities and the loss of degrees of freedom. The task space position vector is augmented by the generalized coordinates of the positioning part. Feedback linearization and decoupling are then applied in the augmented task space to obtain a model consisting of decoupled double integrators. These decoupled double integrators are controlled by the use of linear quadratic optimal control. The controller has been applied to an eight-link hydraulic spray painting robot. The experiments showed that a high bandwidth and a large working area were possible with moderate motor torques.

A controller for redundant manipulators with a small fast manipulator mounted on a positioning part has been developed. The controller distributes the fast motion to the small fast manipulator and the slow gross motion to the positioning part. A position reference is generated on-line to the positioning part to avoid singularities and the loss of degrees of freedom. The task-space position vector is augmented by the generalized coordinates of the positioning part. Feedback linearization and decoupling are then applied in the augmented task space to obtain a model consisting of decoupled double integrators. These decoupled double integrators are controlled by the use of linear quadratic optimal control. In the optimal control problem the performance index is chosen so that the task-space position reference is tracked with a high bandwidth while the reference to the positioning part is tracked with a low bandwidth. The controller has been applied to a simple planar redundant manipulator and an eight-link spray painting robot in simulation experiments. These simulations showed that a high bandwidth was possible with moderate actuator torques.

In this paper, the robustness of the computed torque technique for manipulator control is investigated in the presence of model errors. The robustness analysis is performed in the frequency domain by means of the block Gerschgorin theorem. The block Gerschgorin theorem gives inclusion regions for the eigenvalues of the linearized state space model. The stability of the system close to a trajectory can then be investigated. The results indicate that relatively large errors in the nonlinear feedback compensation of system nonlinearities may be tolerated without affecting system stability. However, even small errors in the computed inertia matrix may result in instability.