Research map / Vision-language and multimodal / Robot action and navigation
RoboTALES: Learning Reasoning-Guided Robot Policies via Task-Aligned Simulated Futures
Hanan Gani, Tejal Kulkarni, Madhoolika Chodavarapu, Nicklas Hansen, Manmohan Chandraker
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 |
|---|---|---|---|
| train video model with planner conditioning and add VLM critic (ablation) | RoboTALES · RoboCasa (3 tasks) | success rate | Improved |
| add planner sub-goals to unadapted video model at inference only (ablation) | VideoPolicy · RoboCasa (3 tasks) | success rate | Hurt |
| train video model with planner conditioning (no critic) (ablation) | RoboTALES · RoboCasa (3 tasks) | success rate | Improved |
| add VLM critic policy optimization (ablation) | RoboTALES · RoboCasa noisy observations sigma 0.06 | success rate | Improved |
| use joint single-stage training instead of stage-wise (ablation) | RoboTALES · RoboCasa (4 tasks) | success rate | Improved |
| use mean-aggregated critic reward instead of single-frame max (ablation) | RoboTALES critic · 360 rollouts | AUROC for success: 0.59 -> 0.72 | Improved |
| use action horizon 16 instead of 8 or 32 (ablation) | RoboTALES · RoboCasa (3 tasks) | success rate | Improved |
| use SBERT critic instead of CycleReward (ablation) | RoboTALES · RoboCasa (6 tasks) | average success rate | Mixed |
| add joint task-aligned training to SVD (ablation) | SVD video model · UCF101 | Frechet distance: 0.6715 -> 0.2437 | Improved |
| add joint task-aligned training to SVD (ablation) | SVD video model · UCF101 | MMD-RBF: 0.2419 -> 0.0631 | Improved |
| use RoboTALES instead of VideoPolicy with 50 demonstrations | video-policy baseline · RoboCasa (4 tasks) | success rate: 46 -> 53 | Improved |
| add planner to VideoPolicy at inference (ablation) | VideoPolicy · RoboCasa (4 tasks) | success rate: 46 -> 34 | Hurt |
Closest papers by meaning
- RoboCerebra: A Large-scale Benchmark for Long-horizon Robotic Manipulation Evaluation (NeurIPS 2025 · similarity 0.89)
- Zero-Shot Robotic Manipulation with Pre-Trained Image-Editing Diffusion Models (ICLR 2024 · similarity 0.89)
- UniDomain: Pretraining a Unified PDDL Domain from Real-World Demonstrations for Generalizable Robot Task Planning (NeurIPS 2025 · similarity 0.88)
- Video Language Planning (ICLR 2024 · similarity 0.88)
- RoboDreamer: Learning Compositional World Models for Robot Imagination (ICML 2024 · similarity 0.88)
- Robot-R1: Reinforcement Learning for Enhanced Embodied Reasoning in Robotics (NeurIPS 2025 · similarity 0.88)
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.
