Research map / Reinforcement learning / Visual and representation RL
Width-based Lookaheads with Learnt Base Policies and Heuristics Over the Atari-2600 Benchmark
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 N-CPL instead of π-IW(1)+ | Atari-2600 planning · Atari-2600, 53-game benchmark | pairwise game wins (average score): 260 (70.1%) vs 159 (42.9%) | Helped |
| Use N-CPL instead of π-HIW | Atari-2600 planning · Atari-2600, 53-game benchmark | pairwise game wins (average score): 260 (70.1%) vs 200 (53.9%) | Helped |
| Use Depth instead of Classic novelty (ablation) | N-CPL · Atari-2600, 53-game benchmark | pairwise game wins (average score): 260 (70.1%) -> 249 (67.1%) | Hurt |
| Disable novelty pruning (ablation) | N-CPL, producing CPL · Atari-2600, 53-game benchmark | pairwise game wins (average score): 260 (70.1%) -> 227 (61.2%) | Hurt |
| Disable novelty pruning (ablation) | N-CPL, producing CPL · Atari-2600 games with sparse meaningful rewards | pairwise game wins (average score): 55 (65.5%) -> 62 (73.8%) | Helped |
| Use Classic instead of Depth novelty (ablation) | RIW · Atari-2600, 53-game benchmark | pairwise game wins (average score): 103 (27.8%) -> 94 (25.3%) | Hurt |
| Use Classic instead of Depth novelty (ablation) | N-CPL · Atari-2600 games with sparse meaningful rewards | pairwise game wins (average score): 47 (56%) -> 55 (65.5%) | Helped |
| Use N-CPL instead of π-IW(1)+ | Atari-2600 planning · Games with branching factor ≥ 10 | pairwise game wins (average score): 28/33 (84.8%) | Helped |
| Use N-CPL instead of π-IW(1)+ | Atari-2600 planning · Games with branching factor < 10 | pairwise game wins (average score): 11/20 (55%) | Mixed |
Closest papers by meaning
- Planning With Pixels in (Almost) Real Time (AAAI 2018 · similarity 0.88)
- Best-First Width Search: Exploration and Exploitation in Classical Planning (AAAI 2017 · similarity 0.88)
- Planning from Pixels in Atari with Learned Symbolic Representations (AAAI 2021 · similarity 0.85)
- General Policies, Representations, and Planning Width (AAAI 2021 · similarity 0.85)
- Efficient Lookahead Encoding and Abstracted Width for Learning General Policies in Classical Planning (NeurIPS 2026 · similarity 0.85)
- Planning and Learning with Adaptive Lookahead (AAAI 2023 · similarity 0.83)
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.