Motion planning

A computer system comprising one or more computers configured to implement the method of planning ego actions for a mobile robot in the presence of at least one dynamic agent, the method comprising:

• searching for an optimal ego action of a set of possible ego actions in multiple search steps, each search step comprising:
• selecting an ego action from the set of possible ego actions,
• selecting an agent behaviour from a set of possible agent behaviours,
• running an online simulation based on the selected ego action and agent behaviour, and thereby determining a possible outcome of the selected agent behaviour and ego action, and
• assigning at least one reward to the selected ego action, based on at least one reward metric applied to the determined outcome, wherein the selection of the ego action in later search steps is biased, based on the determined outcomes of earlier search steps, towards higher reward ego action(s) but the selection of the agent behaviour in the later search steps is biased towards riskier agent behaviour(s), a risky agent behaviour being an agent behaviour that is, according to the earlier search steps, more likely to result in a lower reward outcome for the selected ego action;
• and
• choosing an ego action of the set of possible ego actions, based on the rewards computed in the search steps.

A non-transitory computer readable medium storing a computer program comprising a plurality of sets of computer readable instructions, which when executed by a plurality of processors cause the method of planning ego actions for a mobile robot in the presence of at least one dynamic agent, the method comprising:

• searching for an optimal ego action of a set of possible ego actions in multiple search steps, each search step comprising:
• selecting an ego action from the set of possible ego actions,
• selecting an agent behaviour from a set of possible agent behaviours,
• running an online simulation based on the selected ego action and agent behaviour, and thereby determining a possible outcome of the selected agent behaviour and ego action, and
• assigning at least one reward to the selected ego action, based on at least one reward metric applied to the determined outcome, wherein the selection of the ego action in later search steps is biased, based on the determined outcomes of earlier search steps, towards higher reward ego action(s) but the selection of the agent behaviour in the later search steps is biased towards riskier agent behaviour(s), a risky agent behaviour being an agent behaviour that is, according to the earlier search steps, more likely to result in a lower reward outcome for the selected ego action;
• and
• choosing an ego action of the set of possible ego actions, based on the rewards computed in the search steps 19.