I believe this to be a problem centered on how best design an interface that facilitates exploratory search.
The diversity of content creates challenges in how to best organize the site and makes a straight up and down hierarchical navigation scheme a bad idea. I would recommend instead to rely on faceted search.
I would recommend that you build a dynamic navigation scheme:
Examine your content and identify the properties most relevant to users exploring your site. Your users won't share common gzals - so you have to build different paths to the content. For example, users that are looking for:
- presentations or videos would use the Content Type dimension
- specific topics would use the Topic dimension. This dimension may be hierarchical, meaning that topics can have subtopics and so on.
- content from specific timeframes of your career would use the Publication Date dimension. Users should be able to provide a range of dates to constrain the results.
This approach will accommodate an audience with varying goals/intents/expectations/prior knowledge of the content. It will also provide power users with the ability to combine dimensions. For example, a user could perform a keyword search and constrain results to only presentations about computer vision published in 2014.
Don't try to create a robotics taxonomy from scratch. Instead, stand on the shoulders of giants and repurpose how academic publishers organize robotics content. The "Robotics and Automation" node of the IEEE Taxonomy could provide inspiration (page 57).
Display only the valid dimensional values for the search or navigation state. This is a core piece of faceted navigation. If the user chooses to constrain the search to only presentations, and there were no presentations published in 2013, do not show 2013 as an option for navigation. Do not lead users to dead-ends.
Dynamically order dimensional values based upon their relevancy to the content. Assume a search is done that returns results in five different topics. When displaying the Topic dimension, rank the individual topics based upon how much matching content is in the results. There are a handful of exceptions to this rule, usually around dimensions that apply universally to all content or those used as primary navigation.
Leverage progressive disclosure to handle hierarchical dimensions without overwhelming users. For example, assume a Location dimension that contains States and their major cities. Displaying the entire Location hierarchy would not be helpful. Instead:
- Split State and City dimensions (but make them dependent on each other)
- Only display City if one or more States are selected
The user would first click a state (California) and then have the option to drill down into specific cities. I believe the LinkedIn people search operates in a similar manner.
Two great resources on faceted navigation in general:
Dynamic faceted navigation in decision making using Semantic
Web technology (published in Decision Support Systems Journal)
Daniel Tunkelang's book Faceted Search. Daniel founded Endeca, one of the first major players in e-commerce faceted navigation.