Overview

The player assumes control of Roderick Hero (sometimes styled as “R. Hero”), a one-man rescue team. Miners working in Mount Leone are trapped, and it’s up to Roderick to reach them.

The player is equipped with a backpack-mounted helicopter unit, which allows him to hover and fly, along with a helmet-mounted laser and a limited supply of dynamite. Each level consists of a maze of mine shafts that Roderick must safely navigate in order to reach the miner trapped at the bottom. The backpack has a limited amount of power, so the player must reach the miner before the power supply is exhausted.

Mine shafts may be blocked by cave-ins or magma, which require dynamite to clear. The helmet laser can also destroy cave-ins, but more slowly than dynamite. Unlike a cave-in, magma is lethal when touched. Later levels include walls of magma with openings that alternate between open and closed requiring skillful navigation. The mine shafts are populated by spiders, bats and other unknown creatures that are deadly to the touch; these creatures can be destroyed using the laser or dynamite.

Some deep mines are flooded, forcing players to hover safely above the water. In later levels, monsters strike out from below the water. Some mine sections are illuminated by lanterns. If the lantern is somehow destroyed, the layout of that section becomes invisible. Exploding dynamite lights up the mine for a brief time.

Points are scored for each cave-in cleared and each creature destroyed. When the player reaches the miner, points are awarded for the rescue, along with the amount of power remaining in the backpack and for each remaining stick of dynamite. Extra lives are awarded for every 20,000 points scored.

Description from Wikipedia

State of the Art

Human Starts

Result Method Type Score from
50496.8 RainbowDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
32464.1 A3C FF (4 days) PG Asynchronous Methods for Deep Learning
28889.5 A3C LSTM PG Asynchronous Methods for Deep Learning
28765.8 A3C FF (1 day) PG Asynchronous Methods for Deep Learning
28554.2 DistributionalDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
26345.3 ApeX DQN DQN Distributed Prioritized Experience Replay
25839.4 Human Human Massively Parallel Methods for Deep Reinforcement Learning
20889.9 PERDDQN (rank) DQN Prioritized Experience Replay
20506.4 PERDDQN (prop) DQN Prioritized Experience Replay
15459.2 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
15207.9 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
15150.9 PERDQN (rank) DQN Prioritized Experience Replay
14992.9 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
14892.5 DDQN DQN Deep Reinforcement Learning with Double Q-learning
12952.5 DQN2015 DQN Massively Parallel Methods for Deep Reinforcement Learning
8963.36 GorilaDQN DQN Massively Parallel Methods for Deep Reinforcement Learning
2454.2 NoisyNetDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
1580.3 Random Random Massively Parallel Methods for Deep Reinforcement Learning

No-op Starts

Result Method Type Score from
105929.4 DQfD Imitation Deep Q-Learning from Demonstrations
55887.4 RainbowDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
38874.0 C51 Misc A Distributional Perspective on Reinforcement Learning
35895.0 DuelingDQN DQN Noisy Networks for Exploration
33860.9 DistributionalDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
31655.9 ApeX DQN DQN Distributed Prioritized Experience Replay
31533.0 NoisyNet-DuelingDQN DQN Noisy Networks for Exploration
30826.4 Human Human Dueling Network Architectures for Deep Reinforcement Learning
30791.0 A3C PG Noisy Networks for Exploration
27153.9 PERDDQN (prop) DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
25763 Human Human Human-level control through deep reinforcement learning
23037.7 PERDDQN (rank) DQN Dueling Network Architectures for Deep Reinforcement Learning
22290.1 DuelingPERDDQN DQN Deep Q-Learning from Demonstrations
21036.5 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
20818.2 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
20437.8 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
20357.0 DDQN DQN Deep Reinforcement Learning with Double Q-learning
20130.2 DDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
19950 DQN2015 DQN Human-level control through deep reinforcement learning
15176.0 DQN DQN Noisy Networks for Exploration
14913.87 GorilaDQN DQN Massively Parallel Methods for Deep Reinforcement Learning
14225.2 DDQN+PopArt DQN Learning values across many orders of magnitude
8471.0 NoisyNet-A3C PG Noisy Networks for Exploration
7295 Contingency Misc Human-level control through deep reinforcement learning
6459 Linear Misc Human-level control through deep reinforcement learning
6246.0 NoisyNet-DQN DQN Noisy Networks for Exploration
5053.1 NoisyNetDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
1027 Random Random Human-level control through deep reinforcement learning