Columbia University RoboPIL Lab
The Columbia University RoboPIL (Robot Perception, Interaction, and Learning) Lab, led by PI Yunzhu Li, is an academic robotics research laboratory at Columbia University that focuses on enabling robots to perceive, model, and act on complex physical environments through learned representations, neural dynamics models, diffusion-based policies, and multimodal sensing. The lab publishes research at top venues such as NeurIPS, CoRL, and RSS, with work spanning dexterous manipulation, 3D scene representations, language-conditioned control, and deformable material handling.
“FlexiTac is an open-source, $30-per-unit tactile sensing system that can be bolted onto virtually any commercial gripper in minutes — and it's already been validated in sim-to-real pipelines, cross-embodiment transfer, and VLA-style learning stacks.”
Source→“a complete FlexiTac unit — sensor pad plus readout electronics — costs approximately $30 at small volumes, with the pad pair (FPCs) running $3.55 at 30-unit volumes and dropping to $1.36 at 1,000-unit volumes”
Source→“fabrication 'can typically be completed within ~5 minutes per pad' and uses a desktop cutting machine — a Silhouette Cameo 5 — to cut components with consistent dimensions”
Source→“FlexiTac has already been deployed on the Robotiq 2F-140, xArm grippers, the ALOHA bimanual system, and the LeRobot gripper — all 'without requiring significant mechanical redesign'”
Source→“'an xArm-based robotic platform equipped with a FlexiTac-instrumented gripper' (Section 3.2)”
Source→“we mount FlexiTac on several common manipulation platforms, including the Robotiq 2F-140, xArm grippers, the ALOHA system, and the LeRobot gripper, without requiring significant mechanical redesign”
Source→“the LeRobot gripper (Hugging Face's open-source robotics platform) are both named as FlexiTac deployment targets”
Source→“Calibration requires tuning only two parameters — normal stiffness k_n and damping k_d — and the paper reports that 'after calibration the normalized tactile histograms closely overlap, which measurably improves downstream fine-tuning stability and reduces sim-to-real degradation'”
Source→“'a portable human data-collection device that captures synchronized visual observations, actions, and dense fingertip tactile signals during natural manipulation'”
Source→“Li's group is consistently publishing across tactile hardware, visuo-tactile learning, and sim-to-real pipelines, suggesting a sustained research agenda in contact-aware manipulation”
Source→“Luo bridges materials science and robotics sensing — a combination increasingly relevant as the field moves toward skin-scale sensing.”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.