In computer graphics there exist mainly two ways to represent 3D shapes: implicit methods and explicit methods. The explicit method is quite obvious which uses triangle meshes to represent any shape precisely, while the distinct disadvantage is hard to compute interactions between triangle meshes which is very important in lots of graphic applications. By contrast, the implicit method is able to estimate the interaction easily but difficult to represent arbitrary shape (especially complex surfaces considered) accurately. In the planning stage of space missions, there is an increasing demand for diverse surface details of small celestial body to be applied in virtual testbed based simulation systems. Implicit surface is one of the most promising solutions to this problem. They are powerful both for the modeling of 3D asteroid models and animating movement of rovers on the virtual testbed. The construction of 3D models comes from basic geometric primitives (\ie, sphere, cone, ...) and incrementally sums up their corresponding scalar fields into more complex shapes which represent shapes easily and compactly. Moreover, this compact representation makes it convenient to compute arbitrary patches of virtual testbed on demand, and meanwhile enables those patches contain a dynamically changing topology. However, one conspicuous weakness in implicit modeling is relying on the manual trial and error method to obtain corresponding parameters of implicit functions, and this work is usually tedious and inefficient. In addition, the implicit modeling system can only generate smooth surface, nevertheless, in most practical applications the surface details are the dominant elements. For instance, in space the terrain-based navigation system and optic-based ground guiding system rely on the terrain features of the celestial surface. Therefore, adding realistic surface details on the implicit surface is another key challenge in the generation of celestial (\ie, asteroid) 3D models. What's more, as the fast iteration of graphics hardware, the demand for high-quality 3D objects in nearly all graphic applications (\ie, AAA games, movies) grows exponentially. The traditional way to create 3D models by artists becomes not only more expensive but also hardly satisfy enormous requirements. Even hiring enough artists to help building the scene, the expense is not sustainable for the 3D industry. In this thesis we propose new methods to automatically generate an implicit representation of 3D asteroid models, inspired not only by sphere packing but also from noise models. They enables: \begin{enumerate} \item[-] a novel invariant shape descriptor to be evaluated on GPU side with CUDA; the statistical histogram of the shape descriptor is used to represent the highly detailed 3D asteroid model, \item[-] an automatic method (AstroGen) to approximate the given constraint shape with sphere packing based metaballs, \item[-] an optimization method which use the distance between different asteroids' histogram as target function and \emph{particle swarm optimization} (PSO) algorithm to optimize the parameters of each asteroid's implicit representation (makes the implicit modeling into a machine learning task), \item[-] a new procedural noise model to generate the surface details on the implicit surface, the details behave in a coherent way with the underlying surface. \end{enumerate} Ever since the arise of general GPU, the computation speed of computers has increased notably faster than its memory bandwidth. The direct consequence of this trend is that compute-intensive algorithms (especially parallelizable algorithms) become increasingly attractive. This is the main reason to explain the recent popularity of procedural methods. We believe that the latest tendency in hardware (\ie, GPU, Cloud Computing) justify the necessity to take a reconsideration of procedural methods. Our procedural algorithm fits this trends quite well and has great potential in nearly all areas of computer graphics.