Research map / Reinforcement learning / Visual and representation RL
Unified Local-Cloud Decision-Making via Reinforcement Learning
Kathakoli Sengupta, Zhongkai Shangguan, Sandesh Bharadwaj, Sanjay Arora, Eshed Ohn-Bar, Renato Mancuso
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace additive reward with the multiplicative reward and add action history (ablation) | UniLCD (raw data) · CARLA Town 10 crowd navigation | ENS (%): 50.04 -> 71.70 | Improved |
| multiplicative reward without history versus additive reward (ablation) | UniLCD (raw data) · CARLA Town 10 crowd navigation | ENS (%): 48.35 -> 48.30 | No clear effect |
| add action history to the routing state (ablation) | UniLCD (embedding) · CARLA Town 10 crowd navigation | ENS (%): 57.20 -> 85.97 | Improved |
| add action history to the additive reward (ablation) | UniLCD (raw data) · CARLA Town 10 crowd navigation | ENS (%): 48.35 -> 50.04 | No clear effect |
| transmit local embeddings to the cloud instead of raw images (ablation) | UniLCD w/ History · CARLA Town 10 crowd navigation | ENS (%): 71.70 -> 85.97 | Improved |
| transmit embedding instead of raw data (ablation) | UniLCD w/ History · CARLA Town 10 crowd navigation | Energy (J/m): 7.83 -> 2.90 | Improved |
| use a larger Stage 2 local backbone instead of Stage 1 (ablation) | UniLCD w/ History · CARLA dense crowd | ENS (%): 85.97 -> 86.78 | Improved |
| use UniLCD routing instead of Deep Sequential RL | UniLCD · CARLA Town 10 crowd navigation | ENS (%): 58.84 -> 85.97 | Improved |
| run all computation in the cloud (Cloud-Only) | Navigation policy · CARLA Town 10 crowd navigation | ENS (%): 63.43 -> 0.00 | Hurt |
| use a tiny local backbone with history (ablation) | UniLCD w/ History (raw data) · CARLA dense crowd | ENS (%): 71.70 -> 0.00 | Hurt |
Closest papers by meaning
- Bridging On-Device and Cloud LLMs for Collaborative Reasoning: A Unified Methodology for Local Routing and Post-Training (ICML 2026 · similarity 0.88)
- Uni-O4: Unifying Online and Offline Deep Reinforcement Learning with Multi-Step On-Policy Optimization (ICLR 2024 · similarity 0.84)
- HybridFlow: Resource-Adaptive Subtask Routing for Efficient Edge-Cloud LLM Inference (ICML 2026 · similarity 0.84)
- Intelligence per Watt: Measuring Intelligence Efficiency of Local AI (NeurIPS 2026 · similarity 0.83)
- NoRD: A Data-Efficient Vision-Language-Action Model that Drives without Reasoning (CVPR 2026 · similarity 0.83)
- Point Cloud Matters: Rethinking the Impact of Different Observation Spaces on Robot Learning (NeurIPS 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
