Can Erdogan

Planar Segmentation of RGBD Images Using Fast Linear Fitting and Markov Chain Monte Carlo

With the advent of affordable RGBD sensors such as the Kinect, the collection of depth and appearance information from a scene has become effortless. However, neither the correct noise model for these sensors, nor a principled methodology for extracting planar segmentations has been developed yet. In this work, we advance the state of art with the following contributions: we correctly model the Kinect sensor data by observing that the data has inherent noise only over the measured disparity values, we formulate plane fitting as a linear least-squares problem that allow us to quickly merge different segments, and we apply an advanced Markov Chain Monte Carlo (MCMC) method, generalized Swendsen-Wang sampling, to efficiently search the space of planar segmentations. We evaluate our plane fitting and surface reconstruction algorithms with simulated and real-world data.

Exploiting symmetries and extrusions for grasping household objects

In this paper we present an approach for creating complete shape representations from a single depth image for robot grasping. We introduce algorithms for completing partial point clouds based on the analysis of symmetry and extrusion patterns in observed shapes. Identified patterns are used to generate a complete mesh of the object, which is, in turn, used for grasp planning. The approach allows robots to predict the shape of objects and include invisible regions into the grasp planning step. We show that the identification of shape patterns, such as extrusions, can be used for fast generation and optimization of grasps. Finally, we present experiments performed with our humanoid robot executing pick-up tasks based on single depth images and discuss the applications and shortcomings of our approach.

Planning in constraint space: Automated design of functional structures

On the path to full autonomy, robotic agents have to learn how to manipulate their environments for their benefit. In particular, the ability to design structures that are functional in overcoming challenges is imperative. The problem of automated design of functional structures (ADFS) addresses the question of whether the objects in the environment can be placed in a useful configuration. In this work, we first make the observation that the ADFS problem represents a class of problems in high dimensional, continuous spaces that can be broken down into simpler subproblems with semantically meaningful actions. Next, we propose a framework where discrete actions that induce constraints can partition the solution space effectively. Subsequently, we solve the original class of problems by searching over the available actions, where the evaluation criteria for the search is the feasibility test of the accumulated constraints. We prove that with a sound feasibility test, our algorithm is complete. Additionally, we argue that a convexity requirement on the constraints leads to significant efficiency gains. Finally, we present successful results to the ADFS problem.

Incorporating kinodynamic constraints in automated design of simple machines

Robots are inherently limited by constraints on their motor power, battery life, and structural rigidity. Using simple machines and exploiting their mechanical advantage can significantly increase the breadth of a robots capabilities. In this work, we present an autonomous planner which allows a robot to determine how arbitrary rigid objects in its environment can be utilized in machine designs to overcome physical challenges. First, the designed structure must be sufficient to achieve a task given the input force and torque that can be applied by the robot. Second, the structure must be accessible to the robot given its kinematics and geometry so that it can actually be used to perform the task. The output of our algorithm is the configuration of the design components, the pose of the robot to make contact with the design, and the motor torques needed to actuate it. We demonstrate results with the robot Golem Krang, using levers as simple machines, to overturn 100 kg load and to push 240 kg wheeled obstacle.

Autonomous Realization of Simple Machines

For robots to become integral parts of human daily experience, they need to be able to utilize the objects in their environment to accomplish any range of tasks. In this work, we focus particularly on physically challenging tasks that push the limits on the robot kinodynamic constraints such as joint limits, joint torques and etc. Previously, we demonstrated an autonomous planner that instructs a human collaborator where to place the available objects in the environment to form a simple machine such as a lever-fulcrum assembly. In this work, we report results on the autonomous realization of such a design by the humanoid robot Golem Krang, focusing on the challenges of autonomous perception, manipulation and control.

Autonomous design of functional structures

We present an algorithm that enables a humanoid robot to reason about its environment and use the available objects to build bridges, stairs and lever-fulcrum systems. Facing a challenge that is otherwise intractable, such as climbing a height or moving a heavy object, the proposed planner reasons about the physical limitations of the robot to design functional structures. Inducing constraints on the space of possible designs within a classical planning framework, the algorithm outputs feasible structures that can be used towards accomplishing goals. We present results in dynamic simulation with Golem Hubo, walking on a bridge to cross a hazardous area, and in real-world with Golem Krang, overturning 100 kg loads and pushing 240 kg obstacles. Graphical Abstract