BT-Agent: Deploying Onboard Robot Agents with Behavior Trees and Multi-Fidelity Evaluation
We propose BT-Agent, an onboard architecture combining LLM-driven task decisions with predefined behavior-tree (BT) skills.
ProofPaper ↗
Key points
- Deploying language-model agents onboard mobile robots requires task reasoning and responsive execution within limited compute and memory.
- These skills enforce execution conditions and return structured failure feedback.
- We demonstrate fully onboard operation with a 9B model on a quadruped across five task categories.
- To compare agent designs efficiently, we introduce a 161-task multi-fidelity benchmark, evaluated in text on all tasks and in Habitat on 39.
Sources (1)
- [1]BT-Agent: Deploying Onboard Robot Agents with Behavior Trees and Multi-Fidelity EvaluationarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 9, 05:08 PM
We propose BT-Agent, an onboard architecture combining LLM-driven task decisions with predefined behavior-tree (BT) skills.
Deploying language-model agents onboard mobile robots requires task reasoning and responsive execution within limited compute and memory.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 9, 2026Probabilistic Ship Trajectory Forecasting using Transformer Models on AIS Data
- Oct 9, 2026Sonic-AMG: Streaming Human Motion Generation from Spatial Audio
- Oct 9, 2026JoyInside-VLM-Agent: Product-Conditioned Multimodal Policy Learning for Embodied Agents
- Oct 9, 2026OGPO: Offline Goal-conditioned Policy Optimization for Recoverable Vision-Language-Action Models
- Oct 8, 2026Reflex: Visually Grounded Reactive Humanoid Control