Research map / Reinforcement learning / Imitation learning
What Matters for Adversarial Imitation Learning?
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| switch from on-policy PPO to off-policy RL algorithm (SAC, TD3, D4PG) (ablation) | AIL agent · 10 continuous-control tasks (Gym and Adroit) | normalized return (95th percentile) | Helped |
| use SAC vs TD3 vs D4PG (ablation) | AIL agent · 10 continuous-control tasks (Gym and Adroit) | normalized return (95th percentile) | No clear effect |
| use AIRL reward instead of -ln(1-D) with RL-generated demonstrations (ablation) | AIL reward function · tasks with RL-generated demonstrations | normalized return (95th percentile) | Helped |
| use -ln(1-D) reward instead of AIRL with human demonstrations (ablation) | AIL reward function · human demonstration tasks (door, hammer) | normalized return (95th percentile) | Helped |
| enable explicit absorbing state (ablation) | AIL · variable-length-episode tasks | normalized return (95th percentile) | Helped |
| use FAIRL reward instead of other rewards (ablation) | AIL reward function · 10 continuous-control tasks (Gym and Adroit) | normalized return (95th percentile) | Hurt |
| replay expert data alongside policy data (ablation) | off-policy AIL agent · 10 continuous-control tasks (Gym and Adroit) | normalized return (95th percentile) | No clear effect |
| pretrain policy with behavioral cloning (ablation) | AIL agent · 10 continuous-control tasks (Gym and Adroit) | normalized return (95th percentile) | No clear effect |
| add AIRL log pi(a|s) logit shift to discriminator (ablation) | discriminator · 10 continuous-control tasks (Gym and Adroit) | normalized return (95th percentile) | Hurt |
| add AIRL reward shaping term to discriminator (ablation) | discriminator · 10 continuous-control tasks (Gym and Adroit) | normalized return (95th percentile) | No clear effect |
| add explicit discriminator regularization (spectral norm, GP, dropout, weight decay) (ablation) | discriminator · easy tasks (HalfCheetah, Hopper, Walker2d, Ant) | normalized return (95th percentile) | No clear effect |
| add explicit discriminator regularization (ablation) | discriminator · harder tasks (Humanoid, Adroit) | normalized return (95th percentile) | Helped |
Closest papers by meaning
- Deterministic and Discriminative Imitation (D2-Imitation): Revisiting Adversarial Imitation for Sample Efficiency (AAAI 2022 · similarity 0.90)
- Unlabeled Imperfect Demonstrations in Adversarial Imitation Learning (AAAI 2023 · similarity 0.90)
- A Bayesian Approach to Generative Adversarial Imitation Learning (NeurIPS 2018 · similarity 0.90)
- Latent Wasserstein Adversarial Imitation Learning (ICLR 2026 · similarity 0.90)
- Diffusing States and Matching Scores: A New Framework for Imitation Learning (ICLR 2025 · similarity 0.89)
- Diffusion Imitation from Observation (NeurIPS 2024 · 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.