harpreet (@DataScienceHarp) header
harpreet (@DataScienceHarp) profile picture

hacker-in-residence/ principal devrel engineer. data curation for physical and visual ai systems. in my physical ai era

I ship daily
Joined April 2020
Photo by harpreet (@DataScienceHarp)Photo by harpreet (@DataScienceHarp)Photo by harpreet (@DataScienceHarp)Photo by harpreet (@DataScienceHarp)Photo by harpreet (@DataScienceHarp)Photo by harpreet (@DataScienceHarp)Photo by harpreet (@DataScienceHarp)Photo by harpreet (@DataScienceHarp)Photo by harpreet (@DataScienceHarp)Photo by harpreet (@DataScienceHarp)
a robot can recognize every object in a room and still not know how to open a single drawer knowing there's a cabinet is scene understanding. knowing which handle to grab, which direction to pull, and that it's a drawer not a door is affordance understanding almost no dataset has it in 3D SceneFun3D: 14,800+ functional elements across 710 laser-scanned indoor scenes. every handle, knob, button, and switch labeled with its affordance, motion axis, motion type, and a natural language task description each scene's 3D point cloud with its iPad video recordings in fiftyone. the affordance annotations live in 3D and project into the video frames check it out here huggingface.co/datasets/Voxe…
5
26
237
19,939
a gaussian splat scene of a playground, with a tornado spinning through it. PhysStudio is a collection of physics-based animation projects for 3D Gaussian Splatting scenes, from solvers to rendered demos. it covers lightning, hurricanes, lava, snow, rain, wind, ice melting, and a robot pouring demo, and builds on NVIDIA Warp 1.6. run the solvers and demos on your own splat scenes with each project's run guide. github.com/zju3dv/PhysStudio #ComputerVision #AI
1
4
313
Zip-NeRF's official JAX pipeline leaves approximately 6-7x of training speed on the table. fused-zipnerf replaces it with hand-written CUDA kernels and analytic gradients, without tiny-cuda-nn or PyTorch autograd. quality matches the official implementation, and rendering is approximately 10x faster. it needs an NVIDIA GPU and installs as an extension to the NeRFICG framework. github.com/fhahlbohm/fused-z… #PhysicalAI #ComputerVision
8
484
if you have 10,000 teleop episodes and no plan to watch any of them welcome, this account is for you if you came for prompts, wrong door
2
3
255
8 of the 50 episodes in the original OpenArm cable tie recording are gone here, and the repo says which ones. OpenArm Cable Tie Threading (openarm_umeow LeRobot v3.0) removes episodes 13, 14, 15, 40, 46, 47, 48, and 49. the remaining 42 episodes (47,515 frames) are re-indexed 0 to 41 in their original order. state and action are unchanged, and videos are re-encoded in AV1. use it to train or evaluate bimanual threading policies with LeRobot. huggingface.co/datasets/etha… #PhysicalAI #Robotics
2
271
remove an object from a Gaussian splat and the patch can look right from the camera you edited in and wrong from the side. ComfyUI-SpEdE finds the object in every capture camera with SAM 3.1, deletes its Gaussians, and fits the fill to all cameras at once. it ships 33 ComfyUI nodes and 10 workflows, runs on your own GPU with no cloud calls, and stops when a gate check fails. you can also extract the object to its own PLY or replace it. github.com/Burgstall-labs/Co… #ComputerVision #AI
3
3
51
1,643
a driving model can predict a good trajectory for the wrong reason. KITScenes-LongTail is built to score that mismatch. each rare driving scenario has six-camera surround video, a 5 s future trajectory of 25 waypoints at 5 Hz, and expert reasoning traces in English, Spanish, and Chinese. i parsed it into FiftyOne so you can play all six cameras in sync and rank clips by Qwen3-VL embedding uniqueness to find the rarest. huggingface.co/datasets/Voxe… #PhysicalAI #AutonomousDriving
2
21
848
smell models are trained on small lab datasets, then asked to work in parks, gyms, and dining halls. in New York Smells, 7,000 smell-image pairs cover 3,500 distinct objects across 60 recording sessions, about 70x more objects than existing olfactory datasets. a Cyranose 320 e-nose with 32 sensors sits on a custom 3D-printed rig next to an iPhone and a RealSense depth sensor. i parsed it into FiftyOne so you can flip between the iPhone image, the depth overlay, and the olfaction heatmap of one sample. huggingface.co/datasets/Voxe… #PhysicalAI #ComputerVision
2
10
665
when there's no spoon, humans scoop with something else. ToolDex has scoop demonstrated with 7 distinct tools, stir with 6, and crush with 5, on egg, dough, sand, and sponge. each trial has 8 synchronized RGB-D cameras, and a large fraction ships with hand-optimized 6DoF object poses, per-frame masks, and hand trajectories. use it for policy learning, affordance discovery, and tool-use generalization studies. tooluse-dataset.github.io/ #PhysicalAI #Robotics
1
8
338
'shake the boba' comes with a motion description that reads 'move to the right and to the left, repeating this sequence 3 times'. MotIF-1K has 1,022 robot demonstrations across 13 task categories, each with video, trajectory overlays, task instructions, and motion descriptions. it was built to benchmark and fine-tune VLMs on motion-based success detection. i parsed it into FiftyOne so you can compare the raw video, trajectory overlay, and 16-keyframe storyboard for one demo. huggingface.co/datasets/Voxe… #PhysicalAI #Robotics
5
257
a dual-arm robot needs 72.5 to 86.1 seconds to drop loose change into a piggy bank. AgiBot World 2026 is real-world data from the AgiBot G2 dual-arm humanoid. this 10-episode subset is HG-DAgger collection, with 7 camera streams and per-frame reward, intervened, and succeed signals. i parsed it into FiftyOne so you can scrub the cameras next to the state and action vectors. huggingface.co/datasets/Voxe… #PhysicalAI #Robotics
3
261
you can stop the revolutionary, but you can’t stop the revolution. oss ftw!
Replying to @Da7_Tech
we're not going anywhere!
1
4
318
a policy trained on third-person demos sees a view it won't have at deployment. LingYu Real-World Robotics Dataset has 139.4 hours and 13,263 bimanual episodes across 15 tabletop task families, teleoperated through a VR headset using only the robot's own cameras. head stereo is 3840x1920, with 62-dimensional proprioception and actions, all synced at 45 Hz. it ships in LeRobot v2.1 format for imitation learning and policy adaptation. huggingface.co/datasets/Ling… #PhysicalAI #Robotics
1
2
235
at highway speed, anything beyond 75 m is already your problem. aiMotive Multimodal Dataset has 176 scenes of LiDAR, camera, and radar with 360-degree coverage, shot by day, at night, and in rain. 26,583 annotated frames, 14 object classes, and about 24% of boxes beyond 75 m. i parsed it into FiftyOne so you can scrub the LiDAR point cloud next to camera images with projected boxes. huggingface.co/datasets/Voxe… #PhysicalAI #AutonomousDriving
3
28
1,383
does your aerial segmentation model hold up as the drone climbs? SkyScenes is a synthetic aerial dataset of 33,600 UAV images from the CARLA simulator, with 28 semantic classes, instance masks, and depth. this subset has 280 samples, all ClearNoon at 0° pitch from 15m, 35m, and 60m. i parsed it into FiftyOne so you can scrub the RGB, depth, and segmentation slices of each scene. huggingface.co/datasets/Voxe… #PhysicalAI #ComputerVision
2
3
36
1,364
pedestrian forecasters learn from bird's-eye or static-camera views, so they never see where the walker was looking. EgoTraj has 75 participants from 14 nationalities walking self-chosen routes across sidewalks, crosswalks, and busy streets in Meta Quest Pro headsets. it carries 1024 x 1024 RGB video at 30 fps, 6DoF head pose, per-frame 3D eye gaze vectors, and VLM-generated scene annotations. use it to train first-person trajectory forecasting that conditions on gaze. github.com/yehiahmad/EgoTraj #PhysicalAI #ComputerVision
5
362
the same eight-sensor rig rides on a car, a boat, and a quadruped. OctoSense records stereo RGB and event cameras, thermal, an Ouster OS1-64 LiDAR, IMU, and RTK GPS on one PPS-locked clock. the 371 car sequences alone run 59 hours, and the full release is 382 sequences. i parsed 8 episodes into FiftyOne so you can filter by platform, split, or sensor coverage and scrub every stream together. huggingface.co/datasets/Voxe… #PhysicalAI #Robotics
1
1
14
618
This is clearly the path to scaling ground level 3d globally. It's wild how good LLMs are getting at making plausible 3d reconstructions of the real world. This literally used to be an IARPA (IC world's DARPA) project a few years ago and now it's a frontier model prompt.
Gemini 4 Carbon is out today - spent a bit of time testing its long-horizon agentic coding on a personal experiment: rebuilding ~220m of downtown Los Gatos in Blender (~50 storefronts + 3D interiors), w/ ~zero manual modeling. Quick look under the hood 🧵 (1/5)
20
37
2
399
25,131
at every timestamp, a robotaxi has 9 cameras, 7 LiDARs, and 3 radars to reconcile. kitscenes-multimodal from KIT is Frankfurt urban driving at 10 Hz, labeled with Lanelet2 HD-map polylines, projected LiDAR depth, and the future ego path. i parsed 680 frames from 4 validation scenes into FiftyOne so you can view every camera next to the fused LiDAR and radar cloud. huggingface.co/datasets/Voxe… #PhysicalAI #AutonomousDriving
1
3
44
1,975
exporting a planetary rover's MCAP recordings takes ages, because the script writes to disk constantly to avoid running out of memory. the BASEPROD scripts export the Bardenas Semi-Desert Planetary Rover Dataset, plot weather station data and LIBS spectra, and apply force/torque, bogie offset, and gyro offset corrections. input data is preserved, and corrected files carry CORRECTED in their names. run them on the dataset download linked in the README. the scripts are MIT licensed. github.com/spaceuma/baseprod #PhysicalAI #Robotics
1
209
a highway work zone with a 40 km/h limit and constant traffic gets watched by three LiDARs that sense differently. HighwayScene pairs an Ouster OS0 (rotating time-of-flight), an Aeva Aeries II (FMCW, with radial velocity per point), and a Blickfeld QB2 (solid-state). 30 episodes of about 20 seconds, built for a cross-sensor background subtraction benchmark. i parsed it into FiftyOne so you can play all three point clouds on one timeline and filter by train, val, or test. huggingface.co/datasets/Voxe… #PhysicalAI #ComputerVision
1
7
370