Claims Browser

Latent transitions in existing WAMs are commonly realized with Transformer-based predictors whose inductive structure is centered on token interaction rather than temporal evolution

"Yet latent transitions are commonly realized with Transformer-based predictors whose inductive structure is centered on token interaction rather than temporal evolution."

Generating action plans in natural language limits the plans to a high level and introduces ambiguity in action execution

"generating action plans in natural language tends to limit the generated plans to a high level, introducing ambiguity in action execution"

Multi-modal Large Language Models used for task planning in industrial human-robot collaboration inherently lack an understanding of system states and do not track state transitions, leading to hallucinated actions that deviate from the intended goal

"MM-LLMs inherently lack an understanding of system states and do not track state transitions, often leading to hallucinated actions that deviate from the intended goal"

Common crowd-navigation benchmarks have an evaluation artifact where learned agents can leave the valid domain and bypass dense crowds without explicit boundary constraints

"we observe an evaluation artifact in common crowd-navigation benchmarks: without explicit boundary constraints, learned agents may leave the valid domain and bypass dense crowds"

Using gameplay data to train robot navigation is naïve and will lead to horrible robot behaviors because games explicitly do not care about interactions with NPCs

Game-based robot training approaches ignore human factors in the environment

Existing dexterous manipulation approaches model skills separately with skill-specific constraints, breaking compatibility and continuity required for long-horizon composition

Vision-and-Language Navigation (VLN) benchmark performance jointly reflects visual navigation ability and use of route structure explicitly supplied by task descriptions, rather than pure navigation capability

Tesla's promises about Optimus are based on magical thinking and have inspired unrealistic humanoid robot expectations in the industry

Waymo has navigation reliability issues, failing to correctly identify blocked roads and selecting suboptimal drop-off locations

Waymo is an experiment rather than a reliable taxi service

Spiking neural networks often lag behind state-of-the-art deep learning models in various applications despite offering potential for compressed communication and low-power inference

Waymo driverless vehicles are ubiquitous in SF (visible within 2-3 blocks) while Tesla Robotaxis are rare and require human drivers

Existing inference runtimes designed for request-response serving do not satisfy the runtime contract of embodied deployment requiring multi-rate execution and latency-first batch-1 inference

VLA models suffer from shortcut learning, latching onto spurious correlations rather than true spatial relationships

Existing visual-language-action models and world models struggle with accurate 3D geometry and physically meaningful forecasting for dynamic manipulation

Humanoid robots are basically oversized toys, and media coverage slides into hype when presenting talking points from humanoid companies and investment bankers