Research map / Reinforcement learning / Imitation learning
Deterministic and Discriminative Imitation (D2-Imitation): Revisiting Adversarial Imitation for Sample Efficiency
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 |
|---|---|---|---|
| Use discriminator-based replay-buffer partitioning and off-policy deterministic policy learning instead of adversarial imitation | D2-Imitation · Walker2d, Swimmer, Hopper; 100K environment interactions | normalized episodic return | Helped |
| Use D2-Imitation rather than DAC | D2-Imitation · HalfCheetah with SAC demonstrations | normalized episodic return | No clear effect |
| Use D2-Imitation rather than DAC | D2-Imitation · InvertedPendulum, MountainCarContinuous, BipedalWalker, LunarLanderContinuous; 50K training timesteps | return: 1000.0 vs 1000.0; -0.045 vs -0.085; 308.32 vs 297.58; -34.63 vs -73.49 | Helped |
| Use D2-Imitation rather than DAC | D2-Imitation · InvertedPendulum, MountainCarContinuous, BipedalWalker, LunarLanderContinuous; 50K timesteps | wall clock time (s): 1526.9 vs 2755.0; 1540.9 vs 2937.4; 1720.2 vs 2458.8; 1586.2 vs 2708.4 | Helped |
| Use D2-Imitation rather than ValueDICE across different demonstration-generating policies | D2-Imitation · HalfCheetah; TRPO, DDPG, TD3 and SAC demonstrations | normalized episodic return | Mixed |
| Remove the discriminator and place policy-generated samples in the nil buffer (ablation) | D2-Imitation · SAC demonstrations; 20 and 10 trajectories | normalized episodic return | Hurt |
| Remove the discriminator and use only demonstration-positive and interaction-zero rewards (ablation) | D2-Imitation · SAC demonstrations; 100K interactions | expert performance | Hurt |
Closest papers by meaning
- Planning for Sample Efficient Imitation Learning (NeurIPS 2022 · similarity 0.92)
- Discriminator-Actor-Critic: Addressing Sample Inefficiency and Reward Bias in Adversarial Imitation Learning (ICLR 2019 · similarity 0.92)
- Adversarial Imitation Learning via Boosting (ICLR 2024 · similarity 0.91)
- Diffusing States and Matching Scores: A New Framework for Imitation Learning (ICLR 2025 · similarity 0.90)
- Exploring Gradient Explosion in Generative Adversarial Imitation Learning: A Probabilistic Perspective (AAAI 2024 · similarity 0.90)
- Diffusion Imitation from Observation (NeurIPS 2024 · similarity 0.89)
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.