Knowledge Base Index
robotics
Robotics systems, tools, frameworks, and data collection infrastructure.
| Article | Summary | Updated |
|---|---|---|
| Teleoperation Systems | Open-source teleoperation systems (DexCap, AnyTeleop, OPEN TEACH, Open-TeleVision) for dexterous manipulation data collection | 2026-08-12 |
| Robot Manipulation Datasets | Teleoperation (DROID, BridgeData V2, RoboMIND, AgiBot World), natural demo (UMI, DexWild), and egocentric (EgoDex, EgoVerse, VITRA, Xperience-10M, EgoScale) datasets — taxonomy, selection guide | 2026-08-12 |
| LeRobot Framework | HuggingFace’s end-to-end robot learning platform: hardware interfaces, dataset tools, policy implementations, async inference | 2026-08-12 |
| Simulation Data Collection | Robotics/CV/HRI approaches to human demo collection in simulation: Isaac Lab, MimicGen, ARCap, EgoMimic | 2026-08-12 |
| Visualization Tools | RViz, Foxglove Studio, Rerun.io (ECS model, Chunk/Arrow internals, send_columns ~100x speedup, Blueprint API), LeRobot Dataset Visualizer — capabilities, trade-offs, use cases | 2026-08-12 |
| World Models: 3D Representations | WAM paradigm vs VLA, JEPA family (V-JEPA 2 / Causal-JEPA / ThinkJEPA / Act-JEPA), World-VLA-Loop co-evolving cycle, 3D/4D survey taxonomy, video world models as simulators | 2026-08-12 |
| Meta Quest Teleoperation Stack | PAPRLE, NeuracoreAI, and ETH Unity stacks for Meta Quest controller teleoperation; oculus_reader ADB bridge; Docker/ROS2 installation pitfalls | 2026-08-12 |
| ICRA 2026 Beyond Teleop Workshop | 29 papers across 8 research directions: sim-to-real at scale, human video → policy, alternative collection interfaces, cross-embodiment transfer | 2026-08-12 |
| Patent Landscape: 3D/4D × Robot Manipulation | US patent map of 3D/4D generative methods × robot safety; five confirmed white spaces as of May 2026; claim strategy notes | 2026-08-12 |
| ML Training Infrastructure | NFS+DVC storage, SLURM vs Determined AI scheduling, Isaac Lab+SLURM+Apptainer, RTX 5090 MIG limitation, MLflow+TensorBoard tracking for small robotics research teams | 2026-08-12 |
| Multi-Modal Data Synchronization | 4-problem sync framework (offset/drift/jitter/rate mismatch) for human Physical AI data collection; why NTP fails; clock domain analysis; contact-event timing stakes | 2026-08-12 |
manipulation
Dexterous manipulation techniques, learning methods, and hardware.
| Article | Summary | Updated |
|---|---|---|
| Dexterous Manipulation | Overview of the field: why it is hard, the VLA convergence, task taxonomy, data and generalization bottlenecks | 2026-08-12 |
| Imitation Learning Methods | Behavioral cloning, diffusion policies, flow matching, interactive IL (DAgger/HITL), VLA architectures; VLA-0 (text-as-actions, 94.7%), LBM multi-task pretraining, generative model families (EBMs/Diffusion/AVMs/GANs) | 2026-08-12 |
| Tactile Devices for LfD | OSMO, TacCap, TAG, DOGlove, DEXOP, DexUMI, Wuji Glove — wearable tactile and haptic devices; Type A (sensing) vs. Type B (feedback) distinction | 2026-08-12 |
| Admittance Control vs. Retargeting | Technical comparison of two dexterous hand control architectures for mocap glove teleoperation | 2026-08-12 |
| LfD Data Collection Paradigms | Four paradigms (kinesthetic, teleoperation, natural demo, passive observation), five technical dimensions, five-level task complexity taxonomy; π0.7: annotation density > volume, subgoal conditioning, compositional generalization | 2026-08-12 |
| Grasp Taxonomy and Type-Conditional Generation | Dexonomy pipeline for GRASP taxonomy (31 of 33 types); 10.7k objects, 9.5M grasps; 82.3% real-robot success | 2026-08-12 |
| Imitation Learning Data Quality Metrics | DC-IL (action divergence + transition diversity), EIL (cycle consistency), L2D (preference learning), Re-Mix (domain reweighting) — five data utilization strategies | 2026-08-12 |
| VLA Research Advances: 2026 | RD-VLA test-time scaling, Compression Gap bottleneck principle, action hallucination, memory augmentation, FTP-1 foundation tactile policy, trajectory-level prompt attacks, CBF+flow matching safety — arXiv W07–W32 synthesis | 2026-08-12 |
hri
Human-robot interaction research, workshops, and HITL surveys.
| Article | Summary | Updated |
|---|---|---|
| HRI Research Landscape | Physical vs. social HRI, LLM inflection point, SIA framework, bidirectionality gap; robot-to-human communication taxonomy (visual/haptic/auditory) and co-adaptation in HITL | 2026-08-12 |
| HITL and VLA Surveys | Five major 2023-2026 surveys on human-in-the-loop learning and VLA models: shared diagnosis, gaps, representation alignment | 2026-08-12 |
| VAM-HRI Workshops | VAM-HRI 2024 (21 papers) and 2025 (11 papers): VR/AR for HRI, LfD data quality, industrial deployment trends | 2026-08-12 |
| HRI Safety Taxonomy | Eight HRI safety categories beyond ISO/TS 15066 physical limits: algorithmic, adversarial, alignment, social, communication, bidirectional uncertainty, shared control | 2026-08-12 |
safety
Robot safety research and frameworks.
| Article | Summary | Updated |
|---|---|---|
| World Models for Robot Safety | SafeDreamer, RWM-U, simulation-based safety probing, VLM risk fields, Tex3D adversarial 3D textures, 3D/4D safety gap analysis — world model as attack surface | 2026-08-12 |
claude-code
Claude Code ecosystem, vault architecture, and automation workflows.
| Article | Summary | Updated |
|---|---|---|
| Vault Architecture | Three-layer vault assessment against Karpathy pattern; ai/ naming gaps, interest-map automation gap; two-layer linter design; claude-obsidian comparison | 2026-08-12 |
| Claude Code Skills System | Skills vs. commands frontmatter features; disable-model-invocation, context:fork, dynamic injection; VPS deployment; arXiv digest configuration | 2026-08-12 |
| Loop Engineering for AI Agents | Designing automated agentic loops: 4-condition test, 5-stage structure, 6 building blocks, failure modes, cost per accepted change metric | 2026-08-12 |