Sangwoon Kim

Bio: Sangwoon Kim is an academic researcher from Massachusetts Institute of Technology. The author has contributed to research in topics: Computer science & Step response. The author has an hindex of 2, co-authored 6 publications receiving 11 citations. Previous affiliations of Sangwoon Kim include Mitsubishi Electric Research Laboratories.

Tactile-RL for Insertion: Generalization to Objects of Unknown Geometry

Abstract: Object insertion is a classic contact-rich manipulation task. The task remains challenging, especially when considering general objects of unknown geometry, which significantly limits the ability to understand the contact configuration between the object and the environment. We study the problem of aligning the object and environment with a tactile-based feedback insertion policy. The insertion process is modeled as an episodic policy that iterates between insertion attempts followed by pose corrections. We explore different mechanisms to learn such a policy based on Reinforcement Learning. The key contribution of this paper is to demonstrate that it is possible to learn a tactile insertion policy that generalizes across different object geometries, and an ablation study of the key design choices for the learning agent: 1) the type of learning scheme: supervised vs. reinforcement learning; 2) the type of learning schedule: unguided vs. curriculum learning ; 3) the type of sensing modality: force/torque vs. tactile; and 4) the type of tactile representation: tactile RGB vs. tactile flow. We show that the optimal configuration of the learning agent (RL + curriculum + tactile flow) exposed to 4 training objects yields an closed-loop insertion policy that inserts 4 novel objects with over 85.0% success rate and within 3~4 consecutive attempts. Comparisons between F/T and tactile sensing, shows that while an F/T-based policy learns more efficiently, a tactile-based policy provides better generalization. See supplementary video and results at https://sites.google.com/view/tactileinsertion.

Efficient Tactile Simulation with Differentiability for Robotic Manipulation

Abstract: : Efficient simulation of tactile sensors can unlock new opportunities for learning tactile-based manipulation policies in simulation and then transferring the learned policy to real systems, but fast and reliable simulators for dense tactile normal and shear force fields are still under-explored. We present a novel approach for efficiently simulating both the normal and shear tactile force field covering the entire contact surface with an arbitrary tactile sensor spatial layout. Our simulator also provides analytical gradients of the tactile forces to acceler-ate policy learning. We conduct extensive simulation experiments to showcase our approach and demonstrate successful zero-shot sim-to-real transfer for a high-precision peg-insertion task with high-resolution vision-based GelSlim tactile sensors. The videos and code are available at: http://tactilesim.csail.mit.edu.

Dynamic Control of a Fiber Manufacturing Process using Deep Reinforcement Learning

Abstract: This paper presents a model-free deep reinforcement learning (DRL) approach for controlling a fiber drawing system. The custom DRL-based control system predictively regulates fiber diameter and produces a fiber with a desired, constant or non-constant, diameter trajectory, i.e. diameter variation along the fiber length. Physical models of the system are not used. The system was trained and tested on a small-scale fiber drawing system, which has non-linear delayed dynamics and stochastic behaviors. For a reference trajectory with random step changes, after 1 hour of training, the DRL controller showed the same root mean squared error (RMSE) as an optimized PI controller; after 3 hours of training, it achieved the performance of a quadratic dynamic matrix controller (QDMC). While the PI feedback controller showed 3.5 seconds of time lag in a step response, the DRL controller showed less than a second of time lag. Controller performance tests on trajectories not used in the training process are conducted; for a sine sweep reference trajectory, the DRL controller maintained an RMSE under 40 m up to a frequency of 45 mHz, compared to 25 mHz for QDMC.

Tactile-RL for Insertion: Generalization to Objects of Unknown Geometry.

Abstract: Object insertion is a classic contact-rich manipulation task. The task remains challenging, especially when considering general objects of unknown geometry, which significantly limits the ability to understand the contact configuration between the object and the environment. We study the problem of aligning the object and environment with a tactile-based feedback insertion policy. The insertion process is modeled as an episodic policy that iterates between insertion attempts followed by pose corrections. We explore different mechanisms to learn such a policy based on Reinforcement Learning. The key contribution of this paper is to demonstrate that it is possible to learn a tactile insertion policy that generalizes across different object geometries, and an ablation study of the key design choices for the learning agent: 1) the type of learning scheme: supervised vs. reinforcement learning; 2) the type of learning schedule: unguided vs. curriculum learning; 3) the type of sensing modality: force/torque (F/T) vs. tactile; and 4) the type of tactile representation: tactile RGB vs. tactile flow. We show that the optimal configuration of the learning agent (RL + curriculum + tactile flow) exposed to 4 training objects yields an insertion policy that inserts 4 novel objects with over 85.0% success rate and within 3~4 attempts. Comparisons between F/T and tactile sensing, shows that while an F/T-based policy learns more efficiently, a tactile-based policy provides better generalization.

Thermally drawn fibers as nerve guidance scaffolds

Abstract: Synthetic neural scaffolds hold promise to eventually replace nerve autografts for tissue repair following peripheral nerve injury. Despite substantial evidence for the influence of scaffold geometry and dimensions on the rate of axonal growth, systematic evaluation of these parameters remains a challenge due to limitations in materials processing. We have employed fiber drawing to engineer a wide spectrum of polymer-based neural scaffolds with varied geometries and core sizes. Using isolated whole dorsal root ganglia as an in vitro model system we have identified key features enhancing nerve growth within these fiber scaffolds. Our approach enabled straightforward integration of microscopic topography at the scale of nerve fascicles within the scaffold cores, which led to accelerated Schwann cell migration, as well as neurite growth and alignment. Our findings indicate that fiber drawing provides a scalable and versatile strategy for producing nerve guidance channels capable of controlling direction and accelerating the rate of axonal growth.

Perturbation based privacy preserving slope one predictors for collaborative filtering

Abstract: The prediction of the rating that a user is likely to give to an item, can be derived from the ratings of other items given by other users, through collaborative filtering (CF). However, CF raises concerns about the privacy of the individual user’s rating data. To deal with this, several privacy-preserving CF schemes have been proposed. However, they are all limited either in terms of efficiency or privacy when deployed on the cloud. Due to its simplicity, Lemire and MacLachlan’s weighted Slope One predictor is very well suited to the cloud. Our key insight is that, the Slope One predictor, being an invertible affine transformation, is robust to certain types of noise. We exploit this fact to propose a random perturbation based privacy preserving collaborative filtering scheme. Our evaluation shows that the proposed scheme is both efficient and preserves privacy.

Robust Pivoting: Exploiting Frictional Stability Using Bilevel Optimization

Abstract: Generalizable manipulation requires that robots be able to interact with novel objects and environment. This requirement makes manipulation extremely challenging as a robot has to reason about complex frictional interaction with uncertainty in physical properties of the object. In this paper, we study robust optimization for control of pivoting manipulation in the presence of uncertainties. We present insights about how friction can be exploited to compensate for the inaccuracies in the estimates of the physical properties during manipulation. In particular, we derive analytical expressions for stability margin provided by friction during pivoting manipulation. This margin is then used in a bilevel trajectory optimization algorithm to design a controller that maximizes this stability margin to provide robustness against uncertainty in physical properties of the object. We demonstrate our proposed method using a 6 DoF manipulator for manipulating several different objects.

FlowBot3D: Learning 3D Articulation Flow to Manipulate Articulated Objects

Abstract: We explore a novel method to perceive and manipulate 3D articulated objects that generalizes to enable a robot to articulate unseen classes of objects. We propose a vision-based system that learns to predict the potential motions of the parts of a variety of articulated objects to guide downstream motion planning of the system to articulate the objects. To predict the object motions, we train a neural network to output a dense vector field representing the point-wise motion direction of the points in the point cloud under articulation. We then deploy an analytical motion planner based on this vector field to achieve a policy that yields maximum articulation. We train the vision system entirely in simulation, and we demonstrate the capability of our system to generalize to unseen object instances and novel categories in both simulation and the real world, deploying our policy on a Sawyer robot with no finetuning. Results show that our system achieves state-of-the-art performance in both simulated and real-world experiments.

Tactile Pose Estimation and Policy Learning for Unknown Object Manipulation

Abstract: Object pose estimation methods allow finding locations of objects in unstructured environments. This is a highly desired skill for autonomous robot manipulation as robots need to estimate the precise poses of the objects in order to manipulate them. In this paper, we investigate the problems of tactile pose estimation and manipulation for category-level objects. Our proposed method uses a Bayes filter with a learned tactile observation model and a deterministic motion model. Later, we train policies using deep reinforcement learning where the agents use the belief estimation from the Bayes filter. Our models are trained in simulation and transferred to the real world. We analyze the reliability and the performance of our framework through a series of simulated and real-world experiments and compare our method to the baseline work. Our results show that the learned tactile observation model can localize the pose of novel objects at 2-mm and 1-degree resolution for position and orientation, respectively. Furthermore, we experiment on a bottle opening task where the gripper needs to reach the desired grasp state.