Research map / Reinforcement learning / Policy gradient and actor-critic
R2P2: A ReparameteRized Pushforward Policy for Diverse, Precise Generative Path Forecasting
Nicholas Rhinehart, Kris M. Kitani, Paul Vernaza
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 |
|---|---|---|---|
| add prior-cost term beta=0.1 (H(q,p~)) to training objective (ablation) | R2P2 RNN · CaliForecasting (K=12) | Test -H(q_theta, p~): -125.0 -> -119.0 | Improved |
| add prior-cost term beta=0.1 (ablation) | R2P2 Field · CaliForecasting (K=12) | Test -H(q_theta, p~): -142.5 -> -132.8 | Improved |
| add prior-cost term beta=0.1 (ablation) | R2P2 Linear · CaliForecasting (K=12) | Test -H(q_theta, p~): -144.5 -> -134.1 | Improved |
| add prior-cost term beta=0.1 (ablation) | R2P2 RNN · CaliForecasting (K=12) | Test -H(p, q_theta): 70.20 -> 66.89 | Hurt |
| add prior-cost term beta=0.1 (ablation) | R2P2 Linear · CaliForecasting (K=12) | Test -H(p, q_theta): 64.02 -> 61.57 | Hurt |
| add prior-cost term beta=0.3 (ablation) | R2P2 RNN · Kitti (K=12) | Test minMSD: 1.574 -> 1.282 | Improved |
| use R2P2 reparameterized pushforward objective instead of standard GAIL | GAIL-WG · CaliForecasting (K=12) | Test -H(p, q_theta): 27.43 -> 45.55 | Improved |
| use R2P2 reparameterized pushforward objective instead of standard GAIL | GAIL-WG · Kitti (K=12) | Test -H(p, q_theta): 39.53 -> 47.45 | Improved |
| use R2P2 RNN instead of CVAE | CVAE · CaliForecasting (K=12) | Test minMSD: 1.680 -> 1.530 | Hurt |
| use CNN discriminator instead of MLP discriminator (ablation) | R2P2 GAIL-WG · CaliForecasting (K=12) | Test -H(p, q_theta): 45.55 -> 43.55 | Hurt |
| use RNN policy instead of linear policy (ablation) | R2P2 · CaliForecasting (K=12) | Test -H(p, q_theta): 64.02 -> 70.20 | Improved |
| use R2P2 instead of GAIL in the cross trimodal domain | GAIL · Cross trimodal | Test -H(p, q_theta): 14.10 -> 40.79 | Improved |
Closest papers by meaning
- DriveLaW: Unifying Planning and Video Generation in a Latent Driving World (CVPR 2026 · similarity 0.88)
- End-to-End Trajectory Distribution Prediction Based on Occupancy Grid Maps (CVPR 2022 · similarity 0.87)
- Solving Motion Planning Tasks with a Scalable Generative Model (ECCV 2024 · similarity 0.87)
- SmartPretrain: Model-Agnostic and Dataset-Agnostic Representation Learning for Motion Prediction (ICLR 2025 · similarity 0.86)
- Copilot4D: Learning Unsupervised World Models for Autonomous Driving via Discrete Diffusion (ICLR 2024 · similarity 0.85)
- COME: Adding Scene-Centric Forecasting Control to Occupancy World Model (NeurIPS 2025 · similarity 0.85)
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.
