Line of sight does not mean that the tx is directional - BUT it does help a RX determine direction.
You want directionality in general.
You say this is for a toy,but even then more detail will help. Telling us what you are actually trying to do in more detail will help improve answer quality. eg if this is for something akin to a "follow me" golf cart for use on any golf course you may value a standalone system. If it is for something more like eg a medicine trolley in a rest home or hospital then the ability to provide position beacons or a guidance grid may be attractive.
The technically easiest solution if you have 1 x TX and 1 x RX is a scanning system in either the RX (preferred) or TX. This will probably be mechanical - it can be achieved electronically but probably adds cost and complexity to a basically simple system.
A mechanical scanner canbe extremely simple and would probably be effective in this application.
You could use IR (Infrared) with an approximately omnidirectional TX and use a rotating mechanical scanner to either rotate a photo sensor or a mirror or an aperture such that the direction that the sensor "sees" is progressively swept through 360 degrees. If you know the angle that the swept input is pointing then you can detect the direction where sighnal input is maximum.
The mirror and aperture arrangements have the advantage that the sensor proper can be non rotating.
IR tends to be reflected quite well from surfaces that are not obviously reflective optically and you may find that maximum signal comes from a reflection. Generally I'd expect that as sender and receiver moved that the tx would tend to be the strongest source on average. If it is possible to have the direct path blocked then reflections may cause some interesting results.
GPS: A potentially useful 'high tech' system is to use GPS receivers at TX and RX with the TX sending its coordinates and the RX working out which way top go from the calculated difference in positions. While raw GPS is not liable to be accurate enough for precision homing, as long as TX and RX use the same satellite constellation they are liable to produce a differential accuracy far better than the absolute accuracy. [Using a standard GPS I can plot a path as I drive along a city street, do a U turn at an intersection and return. The plotted path on a map may suggest that I drove on the footpath or wrong side of the street on the outward leg, but the return path will typically be plotted within a metre of the true path relative to the outward path if the same satellites are used. Drive the same course an hour later and the go and return paths will again be close to correct relative to each other - but the path pair may be 5+ metres different than the earlier path pair. The cost of GPS receivers is such that this may be a viable solution in many applications. TX method for passing signals then becomes unconstrained by the need to determine direction from it.
Even eg WiFi can be rendered directional by use of mechanical or electrical aerial systems.
Other methods are possible but knowing more will help avoid long discourses which turn out not to be relevant.
Added:
You now mention a camera.
If camera is sensitive in IR (some are more so than others) then a pulsed IR LED would provide a visual beacon. A separate camera could do likewise - either by scanning as above with fixed sensor plus a rotating mirror or by rotating a mast with camera on or rotating the whole robot. As the device is meant to seek and destro... er photograph, then initial rotational seeking should be acceptable and potentially expected. Once the "beacon" is seen the robot can move to keep it central in its view field. With a suitably wide range zoom lens on the main camera it would often be possible to zoom in on and photograph the target from a substantial percent of your target range, depending on your photographic framing and other requirements.
As above, telling us as much as you are able about what you are trying to achieve greatly helps the answering process. Death by a thousand questions and information dribbles is a common but unproductive way to home on a solution.