harbor run -d rachittshah/railheadharbor run -d rachittshah/railheadRailhead eval set (dev + ood): models play OpenTTD directly; replay-verified rewards, run-off valuation, priced forks, promise-keeping contracts.
harbor run -d rachittshah/railheadRailhead eval set (dev + ood): models play OpenTTD directly; replay-verified rewards, run-off valuation, priced forks, promise-keeping contracts.
harbor run -d rachittshah/railheadRailhead is a reinforcement learning (RL) environment.
In Railhead, a model plays the transport game OpenTTD directly, as the owner of one company.
The model uses the rh command in a shell. It builds roads, rail, airports and docks, buys vehicles, gives orders and manages the loan.
The reward comes only from a replay of the action log of the agent, from a sealed start state, in a new engine process. After the agent stops, the world continues without the agent (run-off). A change to the live game that is not in the action log has no effect on the reward.
ghcr.io/rachittshah/railhead:0.1.1 (linux/amd64 and linux/arm64)deliver (with hidden impossible contracts), deliver_bid, simops, tycoon, arenadev and ood (arctic and tropic climates) in rachittshah/railhead; train in rachittshah/railhead-trainRailhead is a reinforcement learning (RL) environment.
In Railhead, a model plays the transport game OpenTTD directly, as the owner of one company.
The model uses the rh command in a shell. It builds roads, rail, airports and docks, buys vehicles, gives orders and manages the loan.
The reward comes only from a replay of the action log of the agent, from a sealed start state, in a new engine process. After the agent stops, the world continues without the agent (run-off). A change to the live game that is not in the action log has no effect on the reward.
ghcr.io/rachittshah/railhead:0.1.1 (linux/amd64 and linux/arm64)deliver (with hidden impossible contracts), deliver_bid, simops, tycoon, arenadev and ood (arctic and tropic climates) in rachittshah/railhead; train in rachittshah/railhead-train